ROMANCE SYNTAX, SEMANTICS AND L2 ACQUISITION
AMSTERDAM STUDIES IN THE THEORY AND HISTORY OF LINGUISTIC SCIENCE General Editor E. F. KONRAD KOERNER (University of Ottawa) Series IV – CURRENT ISSUES IN LINGUISTIC THEORY
Advisory Editorial Board Raimo Anttila (Los Angeles); Lyle Campbell (Christchurch, N.Z.) Sheila Embleton (Toronto); John E. Joseph (Edinburgh) Manfred Krifka (Berlin); Hans-Heinrich Lieb (Berlin) E. Wyn Roberts (Vancouver, B.C.); Hans-Jürgen Sasse (Köln)
Volume 216
Joaquim Camps and Caroline R. Wiltshire (eds) Romance Syntax, Semantics and L2 Acquisition. Selected papers from the 30th Linguistic Symposium on Romance Languages, Gainesville, Florida, February 2000.
ROMANCE SYNTAX, SEMANTICS AND L2 ACQUISITION SELECTED PAPERS FROM THE 30TH LINGUISTIC SYMPOSIUM ON ROMANCE LANGUAGES Gainesville, Florida, February 2000
Edited by
JOAQUIM CAMPS CAROLINE R. WILTSHIRE University of Florida, Gainesville
JOHN BENJAMINS PUBLISHING COMPANY AMSTERDAM/PHILADELPHIA
8
TM
The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences — Permanence of Paper for Printed Library Materials, ANSI Z39.48-1984.
Library of Congress Cataloging-in-Publication Data Linguistic Symposium on Romance Languages (30th : 2000 : Gainesville, Florida) Romance syntax, semantics and L2 acquisition : selected papers from the 30th Linguistic Symposium on Romance Languages : Gainesville, Florida, February 2000 / edited by Joaquim Camps, Caroline R. Wiltshire. p. cm. -- (Amsterdam studies in the theory and history of linguistic science. Series IV, Current issues in linguistic theory, ISSN 0304-0763 ; v. 216) Includes bibliographical references and index. 1. Romance languages--Congresses. I. Camps, Joaquim. II. Wiltshire, Caroline R., 1963-. III. Title. IV. Series. PC11.L53 2001 440’45--dc21 2001037886 ISBN 90 272 3723 9 (Eur.) / 1 58811 078 8 (US) © 2001 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. • P.O.Box 36224 • 1020 ME Amsterdam • The Netherlands John Benjamins North America • P.O.Box 27519 • Philadelphia PA 19118-0519 • USA
ACKNOWLEDGMENTS
The papers selected for inclusion in this volume originated as presentations at the Thirtieth Annual Meeting of the Linguistic Symposium on Romance Languages (LSRL 30), with parasession on Current Issues in Romance Language Sociolinguistics and Second Language Acquisition, which took place February 24-27, 2000, at the University of Florida, Gainesville, Florida. The editors gratefully acknowledge the scholars listed below for their generous assistance, first, in selecting the papers to be presented at LSRL 30, and later, for reviewing those submitted for inclusion in the selected papers from the conference: Michel Achard, Theresa Antes, Joseph Aoun, Julie Auger, J.-Marc Authier, Ignacio Bosque, Diana Boxer, Barbara Bullock, Andrea Calabrese, Vicki Carstens, J. Clancy Clements, Heles Contreras, Michel deGraf, Anne-Marie DiSciullo, Donca Farkas, Ralph Fasold, Randall Gess, Grant Goodall, Jorge Guitart, Javier Gutiérrez-Rexach, S.J. Hannahs, Galia Hatav, Julia Herschensohn, Larry Horn, José Ignacio Hualde, Haike Jacobs, Richard Janda, Renée Jourdenais, Ellen Kaisse, Richard Kayne, Paula Kempchinsky, Jurgen Klausenburger, Juana M. Liceras, John M. Lipski, Andrew Lynch, Enrique Mallén, Fernando Martínez-Gil, Diane Massam, Gary Miller, Jean-Pierre Montreuil, Richard Morris, Carole Paradis, Ana T. PérezLeroux, David Pharies, Eric Potsdam, Lisa Reed, Lori Repetti, Susana Rivera-Mills, María Luisa Rivero, Yves Roberge, Ana Roca, Nuria S agarra, Mario Saltarelli, Lisa Selkirk, Ester Torrego, Rena Torres Cacoullos, Christina Tortora, Barbara Vance, Co Vet, Daniel Villa, Marie-Thérèse Vinet, Dieter Wanner, Ratree Wayland, Ann Wehmeyer, Lydia White, Karen Zagona, Mary L. Zampini, Maria Luisa Zubizarreta, and Jan-Worter Zwart Additional thanks to Bill Calin and D.Gary Miller for their Outreach lectures, to Jean and Juanita Casagrande for their tireless service on the LSRL 30 organizing committee, to Maritza Bell-Corrales, Jodi Bray, Jodi Nelms, and other UF graduate students in Linguistics and Romance who made the
VI
ACKNOWLEDGMENTS
conference work, and to Jodi Bray, Konrad Koerner, and Anke de Looper for their assistance in editing this volume. Finally, neither the LSRL 30 conference nor this volume would have been possible without the financial support of the following University of Florida units: The Program in Linguistics The Department of Romance Languages and Literatures The College of Liberal Arts and Sciences The Office of Research and Graduate Programs The English Language Institute
Joaquim Camps & Caroline R. Wiltshire Department of Romance Languages and Literatures & Program in Linguistics University of Florida, Gainesville Box 117405 Gainesville, FL 32611-7405 U.S.A. May 18,2001
CONTENTS
Acknowledgments
v
List of contributors
ix
Romance Syntax, Semantics, and L2 Acquisition
1
Joaquim Camps & Caroline Wiltshire A Class of "Lite" Adverbs in French
9
Anne Abeillé & Danièle Godard Adjective Position and Interpretation in L2 French
27
Bruce Anderson French Complex Inversion in the Light of a Minimalist Program
43
Cedric Boeckx Optionality, Presupposition, and WH-in situ in French Cedric Boeckx,
57
Stateva & Arthur Stepanov
Word Order Shift and Natural L2 Acquisition in a Portuguese Creole
73
J. Clancy Clements Unaccusativity and the Impersonal Construction of French
89
Sarah Cummins Shifting the DP Parameter: A Study of Anglophone French L2ers Randall Gess & Julia Herschensohn
105
CONTENTS
VIII
Constraint Demotion and Null-Subjects in Spanish L2 Acquisition
121
Larry LaFond, Rachel Hayes & Rakesh Bhatt Temporal Location of Events and the Distribution of the Romance Counterparts of since-Adverbials
137
Telmo Móia PRO, Movement and Binding in Portuguese
153
Acrisio Pires On Impersonal Reflexives in Romance and Slavic and Semantic Variation
169
María Luisa Rivero The Role of Syntactic Modifications on L2 Oral Comprehension
197
Nuria Sagarra Syntactic Constraints on Access to Lexical Structure: The Case of Ficar
211
Cristina Schmitt A Comparative Semantics for the Subjunctive Mood in Spanish
227
Elisabeth Villalta Index of Terms & Concepts
243
CONTRIBUTORS
Anne Abeillé Université Paris 7 UFRL, case 7003 2 place jussieu 75005 Paris France
[email protected] Bruce Anderson Indiana University Dept. of French & Italian Ballantine Hall 642 Bloomington, IN 47405 USA brcander@indiana. edu Rakesh Bhatt University of Illinois Department of Linguistics 4088 FLB, 707 S. Mathews Urbana, IL 61801 USA rbhatt @uiuc. edu Cedric Boeckx University of Connecticut Department of Linguistics, U-1145 341 Mansfield Rd Storrs, CT 06269-1145 USA
[email protected] J. Clancy Clements Indiana University Dept. of Spanish and Portuguese, Ballantine Hall 844 1020 E. Kirkwood Avenue Bloomington, IN 47405 USA
[email protected]
Sarah Cummins Université Laval Département de langues, linguistique et traduction Cité universitaire Québec G1K 7P4 Canada Sarah.
[email protected] Randall Gess University of Utah Department of Linguistics 255 S. Central Campus Dr., Rm.2328 Salt Lake City, UT 84112-0492 USA
[email protected] Danièle Godard Université Paris 7 UFRL, case 7003 2 place jussieu 75005 Paris France
[email protected] Rachel Hayes University of Arizona Department of Linguistics Douglass Building 200-East Tucson, AZ 85721 USA
[email protected] Julia Herschensohn University of Washington Department of Linguistics Box 354340 Seattle, WA 98195-4340 USA
[email protected]
X
CONTRIBUTORS
Larry L. LaFond University of South Carolina Department of English Columbia, SC 29208 USA
[email protected] Telmo Móia Faculdade de Letras Alameda da Universidade P-1600-214 Lisboa Portugal
[email protected] Acrisio Pires University of Maryland Department of Linguistics 1401 Marie Mount Hall College Park, MD 20742-7505 USA
[email protected] María Luisa Rivero University of Ottawa Department of Linguistics 70 Laurier East P.O. Box 450, Stn. A Ottawa, Ontario KIN 6N5 Canada
[email protected] Nuria Sagarra University of New Mexico Dept. of Spanish and Portuguese 235 Ortega Hall Albuquerque, NM 87131 USA
[email protected]
Cristina Schmitt Michigan State University Department of Linguistics and Languages East Lansing, MI 48824 USA
[email protected] Penka Stateva University of Connecticut Department of Linguistics, U-1145 341 Mansfield Rd Storrs, CT 06269 USA
[email protected] Arthur Stepanov University of Connecticut Department of Linguistics, U-l 145 341 Mansfield Rd Storrs, CT 06269 USA
[email protected] Elisabeth Villalta University of Massachusetts at Amherst Department of Linguistics South College Amherst, MA 1003 USA
[email protected]
We dedicate this volume to Professor Jean Casagrande, one of the organizers of the first LSRL, which was held at the University of Florida, Gainesville, in 1971, to honor his many contributions to Romance linguistics and the linguistics community at the University of Florida and around the world.
ROMANCE SYNTAX, SEMANTICS, AND L2 ACQUISITION JOAQUIM CAMPS & CAROLINE R. WILTSHIRE University of Florida
0. Introduction The chapters in this volume discuss issues in syntax, semantics, their interface, and how syntactic and semantic considerations affect the process of second language acquisition. This collection of papers illustrates the richness in the field of Romance linguistics and the important contributions of crosslinguistic research and multi-modular approaches to the development of linguistic theory. Among the topics discussed in this collection we find movement, parametric variation in the Determiner Phrase (DP) and issues of word order inside and outside the DP (both from theoretical and second language acquisition viewpoints), interpretation of adverbials, syntactic constraints on lexical structure, a semantic analysis of the subjunctive, and semantic interpretations of impersonal constructions and impersonal reflexive pronouns. 1. Movement and related issues The topic of movement is discussed in the papers by BOECKX, STATEVA & STEPANOV, BOECKX, and PIRES. BOECKX,STATEVA&STEPANOVstudy French questions and consider both wh-movement and wh-in situ. They argue that these two variants do not involve syntactic optionality (optionality of movement). According to the authors, the fundamental interpretive differences between the two variants do not allow for a unified syntactic account of both types. They criticize several aspects of Boskovic's (1998, 2000) characterization of wh-in situ in French, basing their argument on a semantic factor, presupposition, which is associated with wh-in situ but not with a fronted wh-phrase. They propose that in situ wh-phrases in French are focused, and use morphological traits to establish the difference in presupposition. While the wh-in situ is headed by a silent definite head D°, the fronted wh-phrase is not.
2
JOAQUIM CAMPS & CAROLINE WILTSHIRE
BOECKX analyzes French complex inversion and the problems it causes for Pesetsky & Torrego's (2000) account of T-to-C in English. While BOECKX agrees with Pesetsky & Torrego's conception of Case as uninterpretable ('misplaced') Tense, he notes that their motivation for T-to-C contrasts with recent attempts to eliminate head-movement as feature-driven. In BOECKX's proposal there is no need for Attract Closest X, Minimal Compliance, feature-driven head-movement, or the ECP. He argues that French complex inversion can be explained by adopting the claim that the most salient characteristic of the construction (subject clitic doubling) is a case of interrogative inflection, which makes it possible for T to relate to without actual movement. PIRES discusses issues of control based on his analysis of infinitives in Portuguese. He shows evidence that non-inflected infinitives display interpretive properties of obligatory control ( ), while inflected infinitives display non-obligatory control (NOC) properties. He also argues for a DPstructures in Portuguese, applying his analysis to movement analysis of control by the matrix subject, control by the object and adjunct control. This analysis allows one to dispense with the PRO Theorem and the Control Module to account for OC. PIRES provides evidence from binding to support his analysis of control as a result of movement. He also proposes that what licenses the NOC null subjects of inflected infinitives is the infinitives' inflectional morphology. 2. The Determiner Phrase and second language acquisition Three of the papers in this collection study issues of second language acquisition related to the Determiner Phrase: LAFOND, HAYES & BHATT, GESS & HERSCHENSOHN, and ANDERSON. LAFOND, HAYES & BHATT apply Optimality Theory to the analysis of the acquisition of null-subjects in Spanish as an L2. They argue that L2 grammatical development is modeled by the Constraint Demotion Algorithm (Tesar & Smolensky 1998), which works on mismatches between the grammatical properties of the L2 input and the internalized system. They expand on previous research by analyzing the interaction between syntactic properties and discourse properties in the decision to maintain or omit subject pronouns in Spanish. The data they analyzed was collected from a total of 98 participants divided into five groups (from beginners to native speakers). The participants were asked to choose between sentences with overt or null-subjects to complete 40 dialogues. The results indicate a clear developmental pattern in the acquisition of the correct ranking for the constraint DROPTOPIC; that is,
INTRODUCTION
3
subjects that are topic-connected are dropped. LAFOND, HAYES & BHATT also found overgeneralization of the DROPTOPIC constraint in the intermediate stages, where learners omitted subjects that must be maintained. They show how the Constraint Demotion Algorithm accounts for all the developmental stages evident in the learners' choices of null- or overt subjects. GESS & HERSCHENSOHN examine parametric variation in the Determiner Phrase (DP) between English and French. They use evidence from a crosssectional study of English speakers learning French to support the Full Functional Hypothesis (FFH) over the Structure Building Hypothesis (SBH). Their data was collected using a 30-item written DP production task that was administered to 85 college students at four different levels of French instruction. They analyzed the learners' production in terms of use of the determiner, adjective word order, agreement morphology, and intrinsic gender. GESS & HERSCHENSOHN's results favor the FFH and support the three main assumptions of this hypothesis: functional categories (in this case the determiner) are present from the beginning (Schwartz & Sprouse 1996); the activation of functional categories is not linked to the specification of morphology (Lardière 1998); and syntax is acquired before idiosyncratic lexicon (Herschensohn 2000). Also in the context of French as a second language, ANDERSON studies the interpretation of adjectives depending on the relative word order in the Determiner Phrase. In the generative paradigm, the surface noun + adjective order of Romance is the result of overt movement of the noun to the head of a higher functional category, while the Germanic adjective + noun order results from covert movement. ANDERSON considers two interpretations for adjectives: (a) an intersective interpretation, where two simple predicates are conjoined, as in A gray elephant (an individual that is both an elephant and is gray), and (b) a nonintersective interpretation, as in A small elephant, which refers to an individual (x) that is small in relation to the property of being an elephant. In French the first interpretation is typical of post-N adjectives, while the second one is typical of pre-N adjectives. ANDERSON reports on a two-part acceptability judgment task administered to 21 learners in a third-year French phonetics course. The task was such that the position of the adjective in the two options given for each situation was crucial for interpretation. The results show that learners know that in post-N position a nonintersective interpretation is significantly less acceptable than an intersective interpretation. On the other hand, in pre-N position both interpretations are available. This second point seems to indicate that the learners have not yet adopted a noun-movement analysis of
4
JOAQUIM CAMPS & CAROLINE WILTSHIRE
French. Finally, ANDERSON mentions preliminary results of another study where the modified noun phrase appears in sentence-medial position. These results suggest that learners may indeed have adopted a noun-movement analysis. 3. Word order and other syntactic factors in SLA Word order plays a role in two additional studies on second language acquisition. This role is central in the study by CLEMENTS, while the study by SAGARRA considers word order together with other syntactic factors. CLEMENTS discusses word order shift in the context of a language contact situation involving Korlai, an Indo-Portuguese creole language, and Marathi. CLEMENTS' data shows a shift from head-initial to head-final both for the Verb-Object and the Verb-Adjunct construction and argues that it may be due to the influence of Marathi, a head-final language. The data analyzed consists of oral narratives produced by 16 native speakers of Korlai and six native Marathi speakers. CLEMENTS' analysis allows him to establish four stages based on the differential development for Object and Adjunct order shift relative to the verb. The effect of the first language on the second is also discussed regarding the production of Korlai children who, despite not showing complete transfer of Korlai word order frequencies to Marathi, fail to show native word order in Marathi. In order to be able to account for the phases in the word order shift where there is little discrimination between head-initial and head-final orders, CLEMENTS supports positing a third option for the word order parameter, by adding "head-initial & head-final" to the standard "head-initial" and "headfinal". His arguments include: (1) the fact that Adjunct and Object order shift operate on separate axes, and (2) the existence of verb doubling and preposition doubling in Korlai. This proposal requires that two systems be part of the same grammar of a speaker at the same time. In this case CLEMENTS proposes bidirectional case assignment. In such a system it is possible for a grammar to contain elements that have an indeterminate status. SAGARRA's study examines the role of simplified syntax in oral comprehension of Spanish as a second language. The participants in this study were 382 students in a second semester Spanish course at the university level. They were instructed to listen to a series of sentences, and to write down everything they remembered after each one. Three elements were used (subordination, complexity, and word order) to create four versions of each target sentence: unmodified, no subordination, shorter sentences, and no Object-Verb-Subject (OVS) structure. Every participant was exposed to
INTRODUCTION
5
three examples of each of the four versions in a within-group counterbalanced design. The analysis was based on the number of idea units recalled for each sentence. The results show no significant difference between the version without subordination and the unmodified version. On the other hand, comprehension was significantly better for the shorter sentences and for the sentences without OVS structure than for the unmodified version. The findings of this study help us obtain a clearer picture of what constitutes comprehensible oral input, and enhance our knowledge of the effects of input simplification in second language instruction. 4.
Advertíais Adverbials are the main topic discussed in the contributions by MÓIA, and ABEILLÉ & GODARD. MÓIA analyzes the Romance counterparts of English since-adverbials in their use as 'inclusive temporal adverbials'. He argues that quantification over events may act as a constraint on the distribution of non-punctual temporal adverbials, and proposes a distinction between two different modes of temporal location: simple inclusive location and fullscanning inclusive location. The contribution of the temporal adverbial is different in the two cases. In structures with simple inclusive location it merely locates the described eventuality, while in structures with fullscanning inclusive location it not only locates the described eventuality (a maximal event), and sub-events composing it, but also contributes to defining that eventuality. MÓIA's analysis stems from the fact that inclusive desde ("since") is only compatible with the full-scanning location, in combination with event descriptions. ABEILLÉ & GODARD examine and contrast the syntactic distribution of monomorphematic degree and (verbal) manner adverbs in French with derived or complex adverbs of the same semantic class. They examine two proposals that can be applied to the analysis of these adverbs: the weak/strong distinction made by Cardinaletti & Starke (1994), and the higher/lower classes of adverbs proposed by Cinque (1999). ABEILLÉ & GODARD argue that these two proposals contain inadequacies and discuss their own weight-based theory of word order within the framework of Headdriven Phrase Structure Grammar. They use lexical weight to base the contrast between the two adverb categories mentioned above, with the first category (monomorphematic degree and (verbal) manner adverbs) being lighter than the other (derived or complex adverbs). This lite vs. non-lite distinction is relevant for words as well as phrases. Lexical liteness can be
6
JOAQUIM CAMPS & CAROLINE WILTSHIRE
deduced when monomorphematicity and semantic type are combined. Phrasal liteness occurs when two lite elements appear in coordination, or when a lite element modifies a lite head. Liteness predicts a more constrained distribution with a quasi adjacency to the head, which can be seen as the mirror image of the heaviness constraint (moving the heavy elements as far from the head as possible). The authors claim that their weight-based theory can be extended to both Romance languages and languages from other families. 5. Syntax/Semantics interface The articles by SCHMITT, VILLALTA, CUMMINS and RIVERO show how syntax and semantics can be combined to offer explanations for a variety of structures. SCHMITT discusses syntactic constraints on access to lexical structure and focuses on the verb ficar in Brazilian Portuguese. This verb has two main meanings: stay/remain and become. Most of what is necessary to account for the behavior of ficar depends on the type of complement it takes. Only the stay meaning is possible when the complement is a locative. Become readings are always available with property denoting predicates whereas the stay readings in this case are possible depending on the context. SCHMITT argues that in all cases, ficar is one single verb, which denotes a transition and has one general argument structure and one event structure. The different readings for ficar are explained by the interaction between the transition component of the verb and its complements. SCHMITT proposes that ficar encodes sub-event information but is unable to license a thematic referential argument on its own. Given that ficar is not a causative verb, the author assumes that it is of category V rather than v, and that the second subevent has to be a state. VILLALTA's paper is an attempt to offer a semantic characterization of the licensing contexts for subjunctive verb forms in Spanish. Her main emphasis is on the use of the subjunctive in the complement clause of propositional attitude predicates. VILLALTA argues that emotive factive predicates and desire predicates do not rely on the traditional distinction between realis and irrealis. Instead, she investigates the hypothesis that in Spanish the subjunctive appears in contexts that require comparison of alternative propositions. That is, an embedded proposition requires the subjunctive if it is compared to its contextually available alternatives on a scale introduced by the matrix predicate. Empirical evidence for this proposal comes from degree modification, the propositional attitude concept, and preposed
INTRODUCTION
7
complements. Finally, VILLALTA outlines possible extensions of her analysis to other contexts. CUMMINS examines unaccusative mismatches in the Impersonal Construction (I ) of French and argues against the Unaccusative Hypothesis (UH) (Perlmutter 1978), which claims that there are two classes of intransitive verbs: unergatives and unaccusatives. The problem with the UH is that apparently unergative verbs can be found in apparently unaccusative structures. CUMMINS examines, and rejects, three types of explanation for these unaccusative mismatches: (a) that unergative verbs are exceptional, (b) that the structure of the I is not unaccusative, and (c) that certain classes of unergative verbs experience a semantic shift that allows them to map as unaccusatives. She claims that a system of free-linking and postsyntactic compositional interpretation can account for patterns of verb selection in the I . She proposes a principle of postsyntactic interpretation, the Subject Initiation Principle (SIP), which states that when a sentence depicts an event and its initiation, the initiator must be in subject position. Her compositional approach leads to deriving the interpretation of the I from both the meaning of the construction and the individual semantic contributions of each of its elements. RIVERO discusses the similarities and differences between impersonal reflexives in Romance and Slavic. In her analysis, RIVERO characterizes the Romance and Slavic impersonal as a syntactic Simplex Expression anaphor, or defective pronoun with a formally encoded human feature. Semantically, the impersonal is a human indefinite pronoun with existential force in both Romance and Slavic. Because it is a pronoun, the impersonal can antecede itself and escape the Novelty Condition for indefinites. The Polish impersonal can be disclosed not only by adverbs of quantification, like its Spanish and Italian counterparts, but also by unselected datives. Then the clitic functions as an expletive pronoun that transmits properties to its discloser. RIVERO argues that the contrast between Polish and the Romance languages regarding the impersonal use of the reflexive clitic is due to parametric variation in semantics, and consists of a language specific choice of discloser-binder in Polish. The research presented here not only expands our understanding of Romance languages, but also benefits the development of linguistic theory in two ways. The analysis of Romance languages enriches the study of nonRomance languages by suggesting analyses that can be extended to such languages and provides a testing ground for tools developed by analyses of non-Romance languages. The interface between syntax and semantics, and
8
JOAQUIM CAMPS & CAROLINE WILTSHIRE
their connection to second language acquisition similarly provide new tests for developing linguistic theory and our understanding of Romance languages.
REFERENCES Boskovic, Zeljko. 1998. "LF movement and the minimalist program". Proceedings of NELS 28 ed. by P. Tamanji, and K. Kusumoto, 43-51. University of Massachusetts, Amherst: GLSA. . 2000. "Sometimes in situ, sometimes in SpecCP". Step by Step: Minimalist essays in honor of Howard Lasnik ed. by Roger Martin, David Michaels, and Juan Uriagereka. Cambridge, Mass.: MIT Press. Cardinaletti, Anna & Michael Starke. 1994. "The Typology of Structural Deficiency: on the three grammatical classes". Clitics in the Languages of Europe ed. by H. Riemsdijk. Berlin: Mouton de Gruyter. Cinque, Guglielmo. 1999. Adverbs and Functional Heads; a Cross-linguistic Perspective. Oxford: Oxford University Press. Herschensohn, Julia. 2000. The Second Time Around: Minimalism and L2 acquisition. Amsterdam & Philadelphia: John Benjamins. Lardière, Donna. 1998. "Case and tense in the fossilized steady state". Second Language Research 14.1-26. Perlmutter, David. 1978. "Impersonal passives and the Unaccusativity Hypothesis". Proceedings of the Fourth Annual Meeting of the Berkeley Linguistics Society, 157-189. University of California, Berkeley. Pesetsky, David & Esther Torrego. 2000. "T-to-C Movement: Causes and Consequences". To appear in Ken Hale: A Life in Language ed. by M. Kenstowicz. Cambridge, Mass.: MIT Press. Schwartz, Bonnie & Rex Sprouse. 1996. "L2 cognitive states and the Full Transfer / Full Access model". Second Language Research 12.40-72. Tesar, Bruce & Paul Smolensky. 1998. "Learnability in Optimality Theory". Linguistic Inquiry 29.229-268.
A CLASS OF 'LITE' ADVERBS IN FRENCH* ANNE ABEILLE & DANIELE GODARD IUF, Université Paris 7 & CNRS, Université Paris 7
0.
Introduction It is well known that Romance languages have certain weak or clitic adverbs.1 In this paper, we examine the syntactic properties of a subclass of scalar (beaucoup, 'much') and manner (bien, 'well') adverbs in French, which seems to make them good candidates for being weak forms in the sense of Cardinaletti & Starke (1994) (by contrast with -ment derived strong forms). We show that their ability to be modified or conjoined however casts doubts on such an analysis. After discussing the possible relevance of Cinque's (1999) structural hierarchy, we propose an analysis based on a lite vs non-lite distinction. We formalize our proposal with a feature WEIGHT, which is part of the syntactic description of lexical items as well as of phrases, and is taken into account by constraints on word order and extraction. 1. A class of French adverbs Degree, quantity and (verbal) manner adverbs in French (which we will call Scalar) are known to have the two following syntactic properties: they cannot escape from the VP domain and be in sentence initial position (1), and they can premodify adjectives or adverbs (2).2
* We want to thank for their comments or judgements O. Bonami, F. Corblin, J. Doetjes, J. Jayez, S. Kahane, Kampers, . Molinier, P. Monachesi, F. Namer, the audience at LSRL 30, as well as the French and Dutch members of the PICS research group on adverbs. 1 See for example the clitic forms ben r pur in Italian or mai, si, tot, cam and prea (occuring between the pronominal clitics and the V) in Romanian. 2 Throughout the paper, we provide glosses rather than translations.
ANNE ABEILLE & DANIELE GODARD
10 (1)
a. b.
(2)
Paul travaille trop. "Paul works too much." * Trop Paul travaille. Too-much Paul works trop gentil / trop gentiment "too kind, too kindly"
It has also been observed that they often come in quasi synonymous pairs with -ment adverbs: trop/excessivement ('too much'), assez/suffisamment ('enough'), beaucoup/énormément ('a lot'), peu/modérément ('little'), bien/ correctement ('well'), plus/davantage ('more') etc. In each pair, the first adverb is monomorphematic (from the synchronous point of view)3 or 'simplex', and the second derived (with the suffix -ment) or complex. Contrasting their properties will help us understand this enduring lexical redundancy. The simplex scalar adverbs have the following properties, which contrast with those of derived (or complex) scalar adverbs: (i) If bare, and normally stressed, they occur before the complements (see Blinkenberg 1928, Kayne 1975, Molinier 1990, Kampers 1999):4 (3)
a. b.
(4)
a. b.
Paul va trop au cinéma / * au cinéma trop. "Paul goes too-much to the movies / to the movies too-much." Paul va excessivement au cinéma/ au cinéma excessivement. "Paul goes excessively to the movies / to the movies excessively." Marie comprend bien le cours/ ?? le cours bien. "M. understands well the lesson / the lesson well." Marie comprend correctement le cours/ le cours correctement. "M. understands correctly the lesson / the lesson correctly."
(ii) However, if they are modified, conjoined or bear a focussing stress, they can occur before or after the complements (Blinkenberg 1928):
3
We analyze as monomorphematic certain adverbs that are historically derived: vraiment, extrêmement, relativement (they do not have the 'in an adj manner' paraphrase anymore). 4 They can only follow tout or ça, which we analyze as 'lite' complements (see Abeillé & Godard 1998): Paul fait tout bien /ça bien 'Paul does everything well / this well'
A CLASS OF LITE ADVERBS IN FRENCH
(5)
a.
b.
11
Paul va [vraiment trop] au cinéma /au cinéma [vraiment trop] "P. goes really too-much to the movies / to the movies really too-much." Marie comprend [très bien] le cours /le cours [très bien]. "M. understands very well the lesson / the lesson very well."
(iii) They cannot be extracted, that is, they cannot be clefted: (6)
a. b.
d.
*C'est trop que Paul va au cinéma. "It is too-much that Paul goes to the movies." C'est excessivement que P. va au cinéma. "It is excessively that Paul goes to the movies." * C'est bien que Marie comprend le cours. "It is well that M. understands the lesson." C'est correctement que Marie comprend le cours. "It is correctly that Marie understands the lesson."
(iv) When they modify an adjective which can be prenominal, the modified adjective can be prenominal, which is not the case when the adjective is modified by a -ment adverb (Grevisse 1988, Guimier 1996, Abeillé & Godard 1999 contra Combettes & Tomassone 1988): (7)
a. b.
d.
Une décision trop habile/ Une trop habile décision. "A decision too clever / A too clever decision" Une décision excessivement habile / *Une excessivement habile décision. "A decision excessively clever / An excessively clever decision" Une bien belle fille / Une fille bien belle "A really beautiful girl / A girl really beautiful" Une fille divinement belle / ?? Une divinement belle fille "A divinely beautiful girl / A girl divinely beautiful"
(v) When they modify an infinitival V, they can occur on its left (as well as on its right), which is not always the case for -ment adverbs, and may not be separated from the V by a higher -ment adverb (9):5
(9b) is acceptable if the adverb is parenthetical.
ANNE ABEILLE & DANIELE GODARD
12 (8)
a.
b.
d. (9)
a.
b.
On lui reprochait de (trop) aller (trop) they him blamed for (too-much) going (too-much) au cinéma. to the movies "They blamed him for going to the movies too much." * On lui reprochait d'excessivement aller au cinéma. they him blamed for excessively going to the movies Elle se réjouit de (bien) parler (bien) le roumain she rejoices over (well) talking (well) Romanian "She is happy that she can talk Romanian well." ?? Elle se réjouit de correctement parler le roumain. she rejoices over correctly talking Romanian On lui reprochait de vraisemblablement beaucoup aller they him blamed for likely a lot going au cinéma. to the movies "They blamed him for going likely a lot to the movies." * On lui reprochait de beaucoup vraisemblablement aller au cinéma "They blamed him for a lot likely going to the movies."
The specific properties of simplex scalar adverbs cannot be explained by prosody or morphology only since some of them are polysyllabic {beaucoup, assez) and there are other monomorphematic adverbs which are not so restricted: the locative ici or là, the temporal hier, for example, can scramble with complements or be extracted: (10) a. b.
Il a rangé le livre là / ici / hier "He filed the book there / here / yesterday." C'est ici /là/ hier qu'il a rangé le livre "It is here / there / yesterday that he has filed the book."
2. Two possible approaches and their problems We examine two recent proposals which might be relevant, the weak/strong distinction of Cardinaletti and Starke (1994), and the higher/ lower classes of adverbs of Cinque (1999). While Cardinaletti and Starke mainly examine nominais, they suggest that the weak vs strong distinction is also relevant for adverbs. (Non clitic) weak constituents are similar to strong ones in bearing word stress, but also differ
A CLASS OF LITE ADVERBS IN FRENCH
13
from them in that they can only occur in specific 'derived' positions and cannot be conjoined or modified. Like weak forms, our simplex adverbs cannot be extracted, and the positions where they occur (to the left of the Vinf, before the complements, or to the left of a prenominal adjective) might be said to be 'specific'. However, they fail the other two tests: (i) They can be modified by different adverbs (11). When the modifier is also a simplex scalar adverb, the modified adverb can still modify a prenominal adjective (12a,b), and occur to the left of the Vinf (12c,d): (11) a. Il mange [beaucoup trop] / [dix fois trop]. "He eats much too much / ten times too much." b. Il travaille [bien moins] / [tro is f o is moins]. "He works much less / three times less." (12) a Une [beaucoup trop] importante participation "A much too important turn out." b. Une [bien moins] agréable aventure que prévu "A much less pleasant adventure than foreseen" Il essaie de [bien moins] manger ces derniers temps. " tries to much less eat these days." d. Il souffre de [beaucoup trop] dormir. "He suffers from much too much sleep(ing)." When they are modified by expressions of a different type, the modified adverb behaves like an ordinary adverb, may occur after the complements (14a) and fails to occur to the left of the prenominal adjective (13) and to the left of the Vinf (14b): (13) a.
Une participation [dix fois trop] importante "A turn-out ten times too big" b. *Une [dix fois trop] importante participation "A ten times too big turn-out" (14) a. Il va au cinéma [dix fois trop]. "He goes to the movies ten times too-much." b. ?? Il craignait de [dix fois trop] manger. "He was afraid of eat(ing) ten times too-much"
ANNE ABEILLE & DANIELE GODARD
14
(ii) They can be conjoined with different adverbs (15). When they are conjoined with adverbs of the same type, such coordinations can modify a prenominal adjective (16a,b), as well as occur to the left of the Vinf (16c,d): (15) a. b. (16) a. b. c.
d.
Il travaille toujours [trop ou trop peu] "He works always too-much or too little." Ils participent [plus ou moins] aux séminaires "They attend more or less the seminars." Une [trop ou trop peu] habile décision "A too or not enough clever decision" De [plus ou moins] fortes précipitations (of) more or less heavy rains On lui reprochait sans cesse de [trop ou trop peu] travailler. they him blamed always for too-much or too little work(ing) "They always blamed him for working too much or too little." Ils essaient de [plus ou moins] participer aux séminaires. "They try to more or less attend the seminars."
When they are conjoined with adverbs of a different type, such coordinations behave like complex adverbs, since they cannot modify a prenominal adjective (17b), cannot premodify a Vinf (17d) and can follow the complements (17c): (17) a. b.
d.
Ces idées [trop ou insuffisamment] nouvelles these ideas too or not enough new * Ces [trop ou insuffisamment] nouvelles idées these too or not enough new ideas "These ideas too new or not new enough" Ils participent aux séminaires [trop ou insuffisamment]. "They attend the seminars too-much or not enough." * Ils craignent de [trop ou insuffisamment] participer They are afraid of too-much or not-enough attend(ing) aux séminaires. the seminars
If the inability to be conjoined or modified is part of the definition of weak constituents, it is clear that our simplex adverbs are not weak. Another possibility is to use Cinque's (1999) hierarchy of adverbs. Leaving aside 'circumstantial adverbs' of time, location, and manner, which exhibit a certain ordering freedom, Cinque divides more constrained adverbs
A CLASS OF LITE ADVERBS IN FRENCH
15
into two classes, the higher and the lower adverbs. Analyzing adverbs as functional heads, he proposes that lower adverbs are located lower in the tree than higher ones, and (roughly) separated from them by the (moved) V. However this hierarchy cannot accommodate the properties illustrated above. In this classification, our simplex adverbs are lower adverbs, which comprise all adverbs failing to occur at the beginning of S, and ordered before the complements. But they can also occur to the left of the V (an infinitival).6 When they do, they cannot have wide scope over a conjunction of VP's (Abeillé & Godard 1997). Thus, (18a) does not convey that the adressee must know the lesson well. This is totally unexpected if they are heads (a functional projection) taking the VP as complement: heads are expected to be able to take simple as well as conjoined complements in a general way. (18) a. b.
Tu dois bien apprendre ce cours et le savoir pour demain. "You must learn this lesson well and know it for tomorrow." Paul travaillera vraisemblablement beaucoup cet été. "Paul will probably work a lot this summer."
On the other hand, the property follows if these adverbs are adjoined to the lexical V rather than the VPinf.7 But the adjunct analysis does not fit well in Cinque's (1999) system. If they are always adjuncts to the lexical V and unordered with respect to it (at least with infinitivals), the fact that they can be separated from the V by a higher adverb (such as vraisemblablement in (18b)) is a problem: the latter must also be analyzed as adjuncts to the lexical V rather than functional heads. Adverbs could have two analyses, adjuncts to the left and functional heads to the right of the V, but this goes counter to Cinque's motivation for the hierarchy of functional projections.
6
Note also that the hypothesis that adverbs are strictly ordered functional heads relies on the availability of movement from the head position of the VP to a higher position, since one finds higher adverbs after the V (18b) (which is supposed to have moved), and lower adverbs can occur before or after the Vinf (8a,c). 7 Similarly, the simplex scalar adverbs do not have wide scope over a coordination of APs (contrary to derived adverbs): (i) une fille [trop belle] et contente d'elle / * trop [belle et contente d'elle] (ii) une fille [excessivement belle] et contente d'elle / excessivement [belle et contente d'elle] (i) is not ambiguous and can only mean that the girl is too beautiful on one hand and happy about herself on the other hand, while (ii) is ambiguous between narrow and wide scope of the adverb.
16
ANNE ABEILLE & DANIELE GODARD
Even when they occur in the position expected of lower adverbs, after the V and before the complements, they raise a difficulty. Remember that they acquire an ordering freedom when they are conjoined, modified or stressed (see (5)). To account for these data, one must appeal to movement of the complements or the adverbs. But movement is unmotivated: conjunction or modification of adverbs of the same type cannot induce a type change; as for the complements, how would they know whether the adverb is conjoined or modified? A similar problem arises with the influence of the adverb type on the pre/postnominal position of the adjective. In Cinque's (1994) analysis of the NP, the relative order of the N and the adjective depends on movement of the head N, whose landing site is a function of the semantic type of the A. It is difficult to see how these degree/intensity adverbs could change the semantic type of the A, specially when quasi-synonymous adverbs (the lower adverb and its -ment counterpart) have a different impact (see (7)). To sum up, Cinque's (1999) structural approach encounters empirical difficulties with the positional and scopal properties of our scalar adverbs in the VP, and is clearly inadequate for dealing with adjectival modification in the NP. We can also check that the syntactic contrast between our constrained simplex adverbs and the synonymous derived ones does not come from a categorial or functional difference. One could say, for instance, that the simplex adverbs are degree categories, while the derived ones are true adverbs. But this does not explain why both types can be conjoined (17a,c) and why modification or coordination changes degree words into adverbs (5). Similar data run counter to a functional explanation (according to which the simplex adverbs would be in specifier position while the derived ones would be in adjunct or complement position). 3. The theory of liteness In this section, we show how the weight-based theory of word-order, elaborated in Abeillé & Godard (1997, 1999, 2000) explains the behavior of scalar adverbs. In this approach, the traditional notion of weight (used for postponing constituents such as sentential complements) is revised as having three possible values: lite (for words or phrases specified as such)8, medium for words not specified as lite and most phrases, heavy for the well-known long or complex phrases. Weight is a factor which interacts with other factors
8 We use the term 'lite' instead of 'light' in order to avoid confusion with semantic lightness (such as light verbs).
A CLASS OF LITE ADVERBS IN FRENCH
17
(such as constraints on categories or grammatical functions) to order lighter elements closer to the head and heavier elements further.9 The cross-categorial distinction between lite and non-lite constituents characterizes both words and phrases, and plays a role in combinatory possibilities, word order and extraction. The contrasting properties are the following: - only non-lite constituents can be extracted, - only non-lite constituents can scramble, - only lite constituents can premodify a lexical head, - lite words belong to certain morphological and semantic classes, - only (certain) phrases made up of lite elements can also be lite. In the same way as the heaviness constraint maximizes the distance between heavy phrases and the head, the liteness constraint minimizes the distance between lite constituents and the head. As an example, we examine bare common N as complements of certain verbs in French: they must precede the phrasal complements, whereas proper names are not so constrained (19). When they are specified, conjoined or modified, they have the same scrambling possibilities as other NPs (20).10 (19) a. b.
c. (20) a.
b.
c.
9
Ils rendaient hommage au Président/ *au Président hommage. "They paid tribute to the President." La course a donné soif à Paul/ * à Paul soif. the running has given thirst to Paul/ to Paul thirst "The running made Paul thirsty." Ils rendirent Marie au Président / au Président Marie. "They gave Marie back to the President." Ils rendaient [un hommage appuyé] au Président / au Président [un hommage appuyé]. "They paid an insistent tribute to the President." La course a donné [à la fois faim et soif] à Paul / à Paul [à la fois faim et soif]. "The running made Paul both hungry and thirsty." La course a donné [vraiment très soif] à Paul / à Paul [vraiment très soif] "The running made Paul really very thirsty."
We leave open here the question whether weight-based constraints only order elements not ordered otherwise or may overrule other word order constraints. 10 The order Object + Indirect Object is sometimes more natural without context, for reasons of focus, which we leave aside.
18
ANNE ABEILLE & DANIELE GODARD
Bare common nouns in French are thus lexically specified as lite, while proper names are medium-weight, and NPs medium-weight or heavy like other phrases. Summarizing, lite constituents precede the non-lite complements in the VP while non-lite constituents can scramble (unless they are heavy and thus postposed). Conjunctions and modifications involving lite constituents can be lite or non-lite. Lexically, adverbs are non-lite by default: the simplex scalar adverbs are lite, while all derived -ment adverbs are non-lite. Hence the ordering constraints on simplex scalar adverbs.11 The distinction between the two classes of adverbs also shows up in phrases with the adverb modifying a lexical V or A. Only lite adverbs (or lite AdvPs) can modify a lexical (infinitival) V on its left, and only lite adverbs (or lite AdvPs) can modify a prenominal adjective. Coordination of lite adverbs or modification of a lite adverb by a lite adverb can be lite (see (12), (16)). However, they can also occur after the complements in the VP (see (5)). The data follow if such phrases can be either lite or non-lite. On the other hand, non-lite -ment adverbs either cannot occur to the left of the V (see (8b,d)), or, if they do, they are adjoined to the VP rather than to the lexical V.12 Finally, a lite adverb modified by a non-lite one is non-lite; accordingly, it cannot left-adjoin to the lexical A or V, as illustrated in (13-14). In the NP, we analyze the A as adjoined to the N, to the left for lite adjectives, and to the right for non-lite ones (Abeillé & Godard 1999, 2000). As for adverbs, only a few (monomorphematic) adjectives are lexically specified as lite, those which are clearly perceived as complex are non-lite, the rest being underspecified.13 In addition, lite adjectives are semantically restricted. As with adverbs, the modification of a lite adjective by a lite adverb can be lite (or non-lite), while the modification of an adjective by a non-lite adverb is obligatorily non-lite. Accordingly, the (lite) A modified by a lite adverb can occur to the left of the N (see (7a,c)), while the A modified by a non-lite adverb can only occur to the right of the N (see (7b,d)). In 11 Other monomorphematic adverbs such as negative and quantifier ones (pas 'not', plus 'no more' jamais 'never', toujours 'always' encore 'still') are also specified as lite since they have the same properties in the VP. They differ from scalar adverbs in that they premodify AP or VPinf (rather than lexical As or Vs), since they can have wide scope over coordinations of APs or VPs, and cannot occur with prenominal adjectives. 12 They do have wide scope over a conjunction of VP's. (i) Il en profiterait pour inévitablement sécher les cours et aller au cinéma "He would take advantage of things to naturally play truant and go to the movies." 13 Adjectives which can occur both to the left and to the right of the N with the same meaning are lexically unspecified, and get their weight from the context.
A CLASS OF LITE ADVERBS IN FRENCH
19
addition, a lite A modified by a conjunction of lite adverbs is lite or non-lite, as expected, since such APs can be prenominal (see (12) and (16a,b)).14 Finally, we assume that only non-lite constituents can be extracted. This is not only justified by the behavior of adverbs (6), but also of nouns. Hence the contrast in (21) between lite nouns (lite) and non-lite proper names or NPs: (21) a. b. c.
* C'est hommage qu'ils ont rendu au Président. "It is tribute that they paid to the President." C'est [un hommage appuyé] qu'ils ont rendu au Président "It is a strong tribute that they paid to the President." C'est Marie qu'ils ont rendue au Président. "It is Marie that they gave back to the President."
4. A Formal representation of the analysis We now embed our analysis in the framework of Head-driven Phrase Structure Grammar (Pollard & Sag 1994). To put it briefly, an HPSG grammar is surface-based (it does not condone movement), and espouses strong lexicalism (syntactic rules do not have access to word parts). Formally, there is a domain of typed linguistic objects, where a type can inherit from several, compatible types, and is associated with an appropriate feature structure. To deal with our data, we use a feature WEIGHT, included in syntactic descriptions of words and phrases. It can have three values: lite, middleweight and heavy. Leaving aside heaviness phenomena here, we will speak of lite and non-lite constituents, where non-lite means 'non-lite and non-heavy'. Lexical items are specified as lite, as non-lite or left underspecified, in which case the lexical item description says nothing (no lexical item can be heavy). Monomorphematic scalar adverbs are lexically specified as lite and the derived -ment adverbs as non-lite. We also use three phrasal types: one for the VP, one for the combination of the lexical V and the adverb or of the lexical N and the adjective, and one for coordinated phrases. Assuming a flat structural representation for the French VP, with adverbials analyzed as complements at the same level,15 we 14 The distinction between lite and non-lite phrases cannot be reduced to a distinction between left and right branching structures, proposed in Alexiadou (1997). It may well be that all right branching phrases (with complements or posthead adjuncts) are non-lite but not all left-branching ones are lite (see e.g. excessivement habile in (7) and also (13)-(14)). 15 As in Abeillé & Godard (1997, 2000) and Bouma et al. (2000), we analyze postverbal adverbs as complements, in order to account for their scrambling with complements as well
20
ANNE ABEILLE & DANIELE GODARD
use the standard head-complements-phrase of Pollard & Sag (1994); this licenses the VP, and also the N-bar. It has a lite head daughter (the N or the V, possibly coordinated with or modified by a lite constituent), waiting for its complements (non-head daughters). The second phrase is the (binary) headadjunct-phrase of Pollard & Sag (1994). Adjuncts have a feature MOD whose value is identified with the synsem (SS, syntactic and semantic information) of the constituent they modify (the head).16
The last phrasal type we need is for coordination, which we assume is a non-headed phrase (as in Pollard & Sag 1994).
as for their extraction behavior. Technically, all verbs in French undergo an optional lexical rule adding an unspecified number of adverbs on their COMPS list. When they are adjuncts, adverbs are constrained (by their MOD feature value) to modify only certain categories (Vinf for lite scalar adverbs, S or VP for temporal or locative ones). 16 'nelist means 'non-empty-list', while 'list' is not specified as empty or non-empty. The elements within '<...>' form a list, and a list (X) is a list made of elements which all have the property X. 'v' notes the logical disjunction, and 'F|F' the path that goes in a feature structure to a certain feature value. 'U' notes unification.
A CLASS OF LITE ADVERBS IN FRENCH
21
We can now turn to phrasal weight. All head-nexus-phrases (phrases which are neither head-adjunct-phrases nor coordinated-phrases) are non-lite. For the other two types, we propose the following constraints:
According to (24a), the weight of head-adjunct-phrases is the unification of the values of the daughters, or it is non-lite. Since the weight values can only unify if they are compatible, the phrase can only be lite if both daughters are lite (or unspecified); otherwise, it is non-lite (if both daughters are nonlite, the unification gives non-lite; if one is lite and the other non-lite, union fails, and the only possibility is non-lite). A coordinated-phrase is a list of constituents whose weight is the unification of two values or is non-lite, and the constraint works in the same way as for (24a). With these weight specifications, we can now formalize the linearization constraints mentioned above, starting with the VP domain. (25) Ordering constraints on Head-Complements-phrases (i) Head<X (ii) [lite] < [non-lite, ADV -] (iii) [lite, ADV +] < [lite, ADV -] The first constraint says that the head comes first. The second one orders lite complements before non-lite ones, allowing non-lite adverbials (marked as [ADV +]) to escape from the constraint. Although lite adverbs and nouns must precede non-lite argument NP's and PP's, they can be preceded by a non-lite adverb as in (18b). In fact, since such adverbs are (syntactically) unconstrained, there is no rule needed for them. It is characteristic of lite constituents that they are ordered among themselves. For example, the lite adverb beaucoup must precede the lite N hommage. This is ensured by the third constraint. (26) a.
Il rendait [beaucoup] [hommage] au Président. "He paid a lot of tribute to the President."
ANNE ABEILLE & DANIELE GODARD
22 b.
* Il rendait [hommage] [beaucoup] au Président.
We illustrate our proposal for the VP with the linearizations in (27) all meaning 'goes usually too much to the movies'. The only ungrammatical order is when the lite adverb trop follows the non-lite NP les cinémas, violating (25ii). (27) a.
fréquente H[lite] b. * fréquente H[litel c. fréquente H[lite] d. fréquente
H[lite]
trop habituellement les cinémas [lite, ADV+] [non-lite, ADV-] [non-lite, ADV+] les cinémas habituellement trop [non-lite, ADV+] [non-lite, ADV-] [lite, ADV+] habituellement trop les cinémas [non-lite, ADV+] [lite, ADV+] [non-lite, ADV-] les cinémas vraiment trop [non-lite, ADV-] [non-lite, ADV+]
Turning to head-adjunct-phrases, vraiment trop is made of two lite constituents, the head trop, and the adjunct vraiment. According to (24b), such a phrase is, out of context, either lite or non-lite. In (27d), since it follows the non-lite NP, it is constrained to be non-lite. More examples are the adjunction of adverbs to adjectives, and of adjectives to N. The ordering of the two daughters is given in (28): (28) Ordering constraints on Head-adjunct-phrases (i) Non-head [lite] < Head (ii) Head [lite] < Non-head [non-lite] The first constraint orders a lite adjunct before the head. Thus, the adjuncts vraiment and moins precede the head in [vraiment trop] and [moins travailler], respectively. The second constraint says that a lite head (of any category) must precede a non-lite adjunct. Consider the head-adjunct phrases in (29). The phrase trop habile, being made up of an unspecified head and a lite adjunct is, out of context, lite or non-lite (see (24a)).
A CLASS OF LITE ADVERBS IN FRENCH
23
(29) a.
une
b.
une
c*
une
d.
une
trop habile ADJ[lite] décision H[lite] excessivement habile ADJ[non-lite] décision H[N, lite]
décision
H[lite] trop habile ADJ[non-lite] décision H[lite] excessivement habile ADJ [non-lite]
Following (28), it is lite on the left of the N, and non-lite on the right. On the other hand, the phrase excessivement habile can only be non-lite, because modification by a non-lite adverb gives a non-lite phrase (24a). The ordering constraints (28) force this adjunct to occur on the right of the N. There remains to give the constraint linking liteness and extraction. In the extraction analysis developed in Sag (1997), synsems are divided into different subtypes. The synsem type for extracted elements is a gap, as opposed to canonical synsems which are realized locally. The following implicational constraint on gap synsems forces them to be non-lite, in effect excluding the extraction of all lite constituents: (30) 5.
gap-synsem --> [WEIGHT non-lite]
Conclusion We have presented a cluster of properties exhibited by monomorphematic degree and (verbal) manner adverbs in French, and contrasted them with those of derived or complex adverbs of the same semantic class. Since the contrast cannot be attributed to a categorial, structural or functional difference, we have proposed an account in terms of lexical weight, the first type of adverbs being 'lighter' than the others. Their alleviated weight prevents them from scrambling with complements, or being extracted, and enables them to occur in a restricted preverbal or preadjectival position. But they are not weak adverbs, in the sense of Cardinaletti and Starke (1994), since they can be coordinated or modified in the same position. As part of a Weight-based theory of word order, the lite versus non-lite distinction is relevant both for words and phrases. Far from being arbitrary, lexical liteness can be deduced from the combination of two different properties: monomorphematicity and semantic type, while phrasal liteness (for adverbial phrases) is typical of coordination among lite elements, or modification of a
24
ANNE ABEILLE & DANIELE GODARD
lite head by a lite element. The same distinction extends to adverbs of the same semantic class in other Romance languages, and its validity should be extended to other languages as well, such as Korean or Greek (see the data in Sells 1994, and Alexiadou 1997, respectively).
REFERENCES Abeillé, Anne & Danièle Godard. 1997. "The Syntax of French Negative Adverbs". Negation and Polarity, Syntax and Semantics, 29 ed. by D. Forget et al, 1-27. Amsterdam & Philadelphia: John Benjamins. . 1998. "A Lexical Approach to Quantifier Floating". Lexical and Constructional Aspects of Linguistic Explanation ed. by A. Kathol et al., 81-96, Stanford: CSLI Publications. . 1999. "La Position de l'adjectif en français: le poids des mots". Recherches Linguistiques de Vincennes, 28.9-31. . 2000. "French Word Order and Lexical Weight". The Nature and Function of Syntactic Categories, Syntax and Semantics, 32 ed. by R. Borsley, 325-360. New York: Academic Press. Alexiadou, Artemis. 1997. Adverb Placement: a Case Study in Antisymetric Syntax. Amsterdam & Philadelphia: John Benjamins. Blinkenberg, Andreas. 1928. L'Ordre des mots en français moderne. Copenhagen: Munksgaard. Bouma, Gosse, Robert Malouf & Ivan Sag. 2000. "Satisfying Constraints on Extraction and Adjunction", Natural Language & Linguistic Theory, forthcoming. Cardinaletti, Anna & Michael Starke. 1994. "The Typology of Structural Deficiency: on the three grammatical classes". Clitics in the Languages of Europe ed. by H. Riemsdijk. Berlin: Mouton de Gruyter. Cinque, Guglielmo. 1994. "Evidence for Partial N Movement in the Romance DP". Paths toward Universal Grammar, Studies in honor of Richard Kayne ed by G. Cinque, 85-110. Washington, D.C.: Georgetown University Press. 1999. Adverbs and Functional Heads: a Cross-linguistic Perspective. Oxford: Oxford University Press. Combettes, Bernard & Richard Tomassone. 1978. "L'Adverbe comme constituant du groupe de l'adjectif', Verbum, 2.53-68. Grevisse, Marcel. 1988. Le Bon Usage. Louvain: Duculot, 12th edition . Guimier, Claude. 1996. Les Adverbes du français. Ophrys: Gap.
A CLASS OF LITE ADVERBS IN FRENCH
25
Kampers-Manhe, Brigitte. 1999. "Le Comportement syntaxique des adverbes de manière". La Modification adverbiale ed. by R. Bok et al., Amsterdam: Rodopi. Kayne, Richard. 1975. French Syntax: the Transformational Cycle. Cambridge, Mass.: MIT Press. Lonzi, Luigi, 1991. "Which Adverbs in spec VP?" Rivista di Grammatica Generativa, 15. 141-160. Molinier, Christian. 1990. "Une Classification des adverbes en t". Langue Française, 88. Pollard, Carl & Ivan A. Sag. 1994. Head-Driven Phrase Structure Grammar. Chicago: the University of Chicago Press. Sag, Ivan. 1997. "English Relative Clause Constructions". Journal of Linguistics 33:2.431-484. Sells, Peter. 1994. "Sub-phrasal Syntax in Korean". Language Research, 30.351-386, Seoul.
ADJECTIVE POSITION AND INTERPRETATION IN L2 FRENCH BRUCE ANDERSON Indiana University
0.
Introduction Second language (L2) acquisition research within the generative paradigm has recently focused its attention on the structure of Determiner Phrases (DPs) in interlanguage grammars (Dekydtspotter et al. 1997, Gess & Herschensohn this volume, Hawkins 1998, Parodi et al. 1997, White & Bruhn de Garavito 1999). The present study deals with a particular word order pattern that differentiates Romance from Germanic languages: the relative order of adjective and noun. The surface noun + adjective (NA) order of Romance is posited to be the reflex of overt movement of the noun across an adjoined AP to the head of a higher functional category (Bernstein 1993, Cinque 1994, Valois 1991). Such movement is posited to be covert in Germanic, resulting in the opposite surface order adjective + noun (AN). The L2 research cited above asks, in part, whether learners are able to acquire the appropriate setting of this parameter by examining the free or elicited production rates of NA versus AN order. With respect to adjective position in French, English-speaking learners receive instruction to the effect that adjectives are generally postnominal (e.g., roi français "French king"), but a few are exceptionally prenominal (e.g., vieux roi "old king"), and still others change meaning based on their position (e.g., roi ancien "ancient king" vs. ancien roi "former king"). There are therefore two equally plausible explanations for the finding that learners produce NA order in French: (a) it is the consequence of a UG-constrained parametric search space interacting with available L2 input, or (b) it is the consequence of an artificial grammar created through explicit rules and reinforced through classroom input. Given the plausibility of both explanations, a more interesting question for research on UG-endowed constraints is to examine what intuitions L2 French learners have about the (un)acceptability of a given word order under a given interpretation,
BRUCE ANDERSON
28
particularly when the adjective is not found in the position that it "should be" according to instruction.1 Indeed, of the 205 most frequently used adjectives in written French texts, 170 (or 83%) have been attested in an adnominal position opposite to the one predicted by classroom instruction. When adjectives are grouped in this manner (l)-(3), it is evident that explicit rules do not fully capture the degree of variation one actually finds.2 The numbers in brackets represent the total number of tokens, followed by prenominal tokens, then postnominal tokens. The number in boldface represents potential input from written texts that contradicts explicit rules of adjective position.
1
(1)
Group A: uniquely pre-N adjectives mauvais "bad" [152-145-7] vieux "old" [388-380-8] beau "handsome" [447-414-33] nouveau "new" [221-141-80]
(2)
Group B: variable (change-of-meaning) adjectives cher "dear/expensive" [121-113-8] propre "own/clean" [138-114-24] ancien "former/ancient" [560-419-141] grand "great/tall" [1328-1279-49] pauvre "pitiable/penniless" [206-191-15]
(3)
Group C: uniquely post-N timide "shy" violent "violent" délicieux "delicious" lourd "heavy" étrange "strange"
adjectives [13-3-10] [44-15-29] [32-15-17] [81-37-44] [78-48-30]
Coppieters (1987) was the first L2 study to address this question. He found that native and nonnative speakers' intuitions diverged rather extensively, though Birdsong (1992) raised a number of conceptual and methodological problems with the study. 2 The figures presented in this section are based on a review of Forsgren (1978), Wilmet (1980), and Larsson (1993), whose corpora are derived from written French texts in a variety of genres. As Albert Valdman (p.c.) points out, however, such frequency counts say nothing about range (i.e., that the unexpected pre- or post-N position of the adjective might only - or normally - occur as part of a fixed expression, such as pleine lune 'full moon'). Nevertheless, the frequency figures, supplemented by research using the ARTFL database, suffice to establish a much greater amount of variability than is normally discussed in the classroom.
ADJECTIVE POSITION AND INTERPRETATION
29
Far from being ungrammatical, the alternate position of the adjective triggers a subtle difference in interpretation (to be discussed below; see also Delbecque 1990, Waugh 1977, Wilmet 1981). It is the variability seen in (1)(3) that I will exploit in arguing that nativelike attainment is only possible through (a) an innate, domain-specific knowledge of semantic combination interacting with available positive evidence; and (b) an overt noun-movement analysis. 1. Adjectives at the syntax-semantics interface It has long been noted that sentences containing a modified noun phrase differ in entailment patterns depending on the adjective used. For example, knowing that elephants are animals, the sentence Dumbo is a gray elephant entails that Dumbo is a gray animal. However, the sentence Dumbo is a small elephant does not entail that Dumbo is a small animal. This suggests that adjectives are not of a homogeneous class, but rather can be of (at least) two logical types: simple predicates of type <e,t> and complex predicates of type « e , t > , < e , t » . Corresponding to these two logical types are two distinct compositional rules. The expression gray elephant picks out an individual (x) that is both gray and an elephant. This comes about through conjunction introduction (4), in which two simple predicates of type <e,t> are conjoined. (4)
gray<e,t> Xx [gray'(x)] elephant<e,t> Xx [elephant'(x)] gray elephant<e,t> Xx [gray'(x)
elephant'(x)]
While there is, in theory, nothing that bars the same kind of semantic composition with adjectives such as small, this does not in fact obtain in natural languages (cf. the difference in entailment patterns noted above). Rather, the expression small elephant picks out an individual (x) that is small in relation to a particular property P (the property of being an elephant). This interpretation comes about through functional application (5), in which the adjective - as an expression of type « e , t > , < e , t » - applies to a common noun of type <e,t> to derive an expression of type <e,t>. (5)
small«e,t>,<e,t» λP [λx [small'(P)(x) ( )]] elephant<e,t> λx [elephant'(x)] small elephant<e,t> Xx [small'(λy [elephant'(y)])(x) ^ elephant'(x)]
30
BRUCE ANDERSON
I will henceforth refer to the interpretation derived through conjunction introduction as intersective and the one derived through functional application as nonintersective (also referred to as subsective) following much of the literature on this topic (see Chierchia & McConnell-Ginet 1990 and Heim & Kratzer 1998 for an overview). Following the general discussion in Partee & Rooth (1983) and Partee (1987), I posit that adjectives are entered in the lexicon in their simplest logical type but may shift type in predictable ways. Discourse is such that ethnic and color adjectives rarely deviate from their logical type <e,t> and are almost always postnominal (post-N) in French. Adjectives of size, age, goodness, beauty, etc. rarely deviate from their logical type « e , t > , < e , t » and are almost always prenominal (pre-N) in French. The majority of adjectives regularly shift between these two logical types, as is especially the case with the adjectives in (2): cher<e,t> means "expensive" in post-N position and h r<< ,t>,<e,t» means "dear/beloved" in pre-N position.3 I assume that lexical items within NP determine the extensity of an expression - that is, the set of individuals to whom the expression may refer. Functional projections above NP (such as number, gender, defmiteness) determine to which individual(s) the NP actually refers in context. Elements in the functional domain are therefore clearly presuppositional. It makes sense to view adjectives of type « e , t > , < e , t » (whether lexicalized as such or type-shifted) as being quasi-functional elements on a par with number, gender, defmiteness, etc. since all are sensitive to the local context supplied by the head noun and the external context supplied by previous discourse. They are generated outside the NP in the functional domain. This accords with much of the generative literature on DPs in Romance languages (cf., in particular, Bernstein 1993). In the case of an intersective interpretation, (4) would translate as (6) in the syntax of English. The adjective gray is adjoined to the NP elephant. The number feature [± singular] applies to this expression, indicating whether there is a singular or plural individual (as the case may be). The definite article (of type <<e,t>,e>) applies to this expression, deriving an expression of type e: the unique gray elephant in context. 3
It follows, then, that for an adjective such as gris to appear in pre-N position, it must type shift to «e,t>,<e,t» and its meaning must be construed as context-sensitive (i.e., relative to some property P). Conversely, an adjective such as petit in post-N position must type shift to <e,t> and its meaning must be construed as context-independent (concrete, objective). I assume, following Waugh (1977), that the infrequent pre-N position of gris and post-N position of petit results from infrequent contexts in which such a construal is intended.
ADJECTIVE POSITION AND INTERPRETATION
31
In the case of a nonintersective interpretation, (5) would translate as (7) in the syntax. The adjective is above NP in the functional domain, being adjoined to NumP. As before, [± singular] applies to the NP, indicating whether there is a singular or plural individual, as the case may be. The adjective small applies to this expression, asserting that this individual is small in relation to the property of being an elephant. As before, the definite article applies to this expression, deriving an expression of type e: the unique small elephant in context.
Since movement to check off the number feature on N is posited to be covert in English, both the adjective gray and the adjective small are pre-N. Since noun-movement is overt in French, the equivalent of gray - gris in its basic logical type <e,t> - is post-N; the equivalent of small -petit in its basic logical type « e , t > , < e , t » - is pre-N. In other words, an intersective interpretation is uniquely associated with post-N position in French and a nonintersective interpretation is uniquely associated with pre-N position.
BRUCE ANDERSON
32
The context-sensitive nature of pre-N adjective position is seen most clearly when the modified noun phrase is combined with the definite article le/la, as suggested in Martin (1986). The sentence La police a intercepté le dangereux criminel "the police caught the dangerous criminal" conjures up a context in which there is one noun-referent (one criminal) who happens to be violent (hereafter 'the single noun-referent context'). The opposite order of noun and adjective - le criminel dangereux - conjures up a context in which there is likely more than one criminal but only one who is dangerous (hereafter 'the multiple noun-referent context'). Whereas pre-N position of the adjective is obligatory in either context in English (cf. [8a] and [9a]), it is only possible in a single noun-referent context in French (cf. [8b] and [9b]). (8)
Multiple noun-referent context: a. The police caught [the [dangerous<e,t> criminal]] b. La police a intercepté [le[crimineli[dangereux<e,t>ti]]]
(9)
Single noun-referent context: a. The police caught [the [dangerous«e,t>,<e,t» [criminal]]] b. La police a intercepté [le [dangereux«e,t>,<e,t»criminell [ti]]]
2. Grammatical change in interlanguage grammars Schwartz & Sprouse's (1996) Full Transfer/Full Access model of L2 acquisition proposes that the entirety of the L1 grammar (excluding the phonetic matrices of lexical/morphological items) constitutes the initial state of L2 acquisition. Parametric change is failure-driven in the sense that only input that cannot be accommodated within Interlanguage Grammar 1 will force the adoption of Interlanguage Grammar 2, with each intermediate step fully constrained by the principles of UG. The acquisition scenario under a Full Transfer account would be that an initial assumption is made on the part of the English-speaking learner to the effect that a grammar with covert noun-movement is able to generate the surface word order patterns of French. Positive evidence, in the form of postN adjectives, could not (presumably) be accommodated within such a grammar, thus forcing the parametric change to overt noun-movement. It is my intent in this section to demonstrate that such a parametric change is not as straightforward as it might first seem to be. In particular, I argue that the existence of L2 input featuring post-N adjective position may not be a sufficient trigger for grammatical change. In addition, there are particular problems with both classroom input and explicit evidence (instruction) in
ADJECTIVE POSITION AND INTERPRETATION
33
arriving at the interpretive correlates of adjective position in L2 French. This state of affairs predicts much more difficulty in acquiring the proper L2 representation of noun phrases than has been demonstrated in studies looking at noun-adjective word order in the absence of interpretive effects (cf. Gess & Herschensohn this volume, Hawkins 1998, Parodi et al. 1997, White & Bruhn de Garavito 1999). First, it is not the case that adjectives are only post-N in French and only pre-N in English. Not only is French similar to English in the typical pre-N position of adjectives such as bon "good" and petit "small" (thereby providing contradictory evidence with respect to noun-movement), English is similar to French in that post-N order obtains in a number of cases. These include the obligatory post-N position of heavy APs (10a), the optional postN position of deverbal adjectives (10b), and the so-called small clauses in (lla-c) (cited in Bouldin 1991): (10) a. b. (11) a. b.
We have every size [big enough to fit heavyweight wrestlers]Ap. Quaalude is the most distant (visible) star (visible). We delivered the flowers fresh. Pat painted the house red. The restaurant manager wants the coffee hot.
While one may wish to argue that the cases of post-N adjective position in (11) do not constitute true NP constituents, those in (10) certainly do. The point is that, on the Full Transfer account, an English-speaking learner might not initially analyze cases of NA order in French as being the result of overt noun-movement. In that case, correctly producing NA order in L2 French must be viewed as a necessary but insufficient condition for claiming that an overt noun-movement analysis has been adopted. Rather, a number of English-compatible syntactic constructions might be entertained first, including rightward AP adjunction, reduced relative clauses, small clauses, and so forth.4 Second, classroom instruction portrays the interpretive differences in adjective position as being restricted to only those adjectives that have 4
I am not arguing here that there is no input to trigger the parametric change to nounmovement. Nominal constructions such as l'invasion allemande de la Pologne 'the German invasion of Poland' might constitute just such a trigger, since the nominal invasion could (presumably) only be separated from its subcategorized theme de la Pologne as a result of overt movement across the agentive adjective allemande of type <e,t>.
34
BRUCE ANDERSON
distinct lexicalized meanings in the form of English glosses: cher in post-N position means "expensive" and in pre-N position means "dear". It is unclear how such arbitrary lexical correspondences could aid the learner in figuring out interpretive differences when there is no readily identifiable English gloss. Third, it is questionable whether the classroom setting provides the right quality and quantity of input. During the spring semester of 1999, ten hours of in-class observations were conducted at each of six levels of French courses taught at Indiana University. Every instance of adjective use by the instructor and students in both oral and written input was recorded. The findings from this observation period indicate that predicate and post-N positions constitute roughly 75% of the available input. Pre-N position constitutes roughly 25% (as low as 14% but no higher than 27%). Moreover, this 75-25 ratio remains constant across levels. The 25% of pre-N tokens, however, involve only 15 to 41 adjectives, depending on level. That is, only a relatively small group of adjectives appear in pre-N position in classroom input, most of these being ones that learners already expect to appear in this position. As for the number of adjectives that actually varied in position at least once during the observation period, there were none attested at the earliest levels, and this figure only increased to 20 adjectives by the fourthyear lecture course. What is worse, a small portion of this input, in the form of oral input to students from other students, was degenerate (e.g., typically post-N adjectives being placed in pre-N position in an infelicitous context). The learnability problem is in fact much more profound. With respect to post-N position, the only input indicating that a nonintersective interpretation is not also possible simply does not exist. This would seemingly require UGgoverned constraints on the hypothesis space to the effect that NPs having the order noun+adjective be interpreted intersectively (i.e., as the result of conjunction introduction). Withrespectto examples of pre-N position such as le dangereux criminel, the learner must somehow come to know that this position is only acceptable in a single noun-referent context, yet there is no instruction to this effect and nothing in the input could demonstrate that this must be the case: Post-N position in such a context is equally possible.5 Given the various syntactic analyses that can be applied to the L2 input discussed above, as well as the learnability problems associated with such input, the structures under consideration constitute an excellent case of L2 By "equally possible" I mean that both pre- and post-N position in a single noun-referent context results in the same truth-value: le dangereux criminel requires that there be only one criminal (who is violent); le criminel dangereux requires that there be at least one individual.
ADJECTIVE POSITION AND INTERPRETATION
35
word order patterns whose acquisition would require both implicit knowledge of the compositional rules underlying interpretation at the syntax-semantics interface and a noun-movement analysis of DPs. Two specific predictions can therefore be made: (a) a nonintersective interpretation in post-N position should be severely degraded in comparison to an intersective interpretation; and (b) an intersective interpretation should be severely degraded in pre-N position if a noun-movement analysis of DPs has been adopted (both interpretations should be possible if it has not). 3. Study methodology and results To test these two predictions, a two-part acceptability judgment task was administered in a controlled setting to 21 learners of intermediate French enrolled in a third-year course on French phonetics. Learners were asked to read a short paragraph in English followed by a test sentence in French. The logic behind the use of English-language contexts was to ensure that participants fully understood the previous discourse upon which the task sentence was to be judged. One might wish to argue that the use of Englishlanguage contexts biases (or at least encourages) a translation strategy in evaluating the task sentences. However, if learners were doing just that, they should reject the adjectives used in this task in post-N position (light APs and nondeverbal adjectives being ungrammatical in this position in English). Fourteen adjectives, broken down into three groups according to instruction (cf. [l]-[3] above), were featured once prenominally and once postnominally in the same sentence within a context forcing one or the other interpretation (see [12] and [13] for examples). All tokens, as well as distracter sentences, were randomized across the two tasks. Learners were asked to indicate whether the test sentence accurately described the situation in the context by clicking on buttons on the computer screen marked true, false, or cannot decide. The design of the study is such that any interpretive differences could only be attributed to the position of the adjective.
BRUCE ANDERSON
36 (12)
[multiple noun-referent context] Monsieur Marat is the judge at a community baking contest. He's really looking forward to this contest because there are four chocolate cakes to taste (=goûter) - his favorite kind of dessert. Everyone gathers around to watch the judging. The first cake turns out to be too dry; the second is burnt on the bottom; the third is too bland. Luckily, the fourth cake was perfect it practically melted in his mouth. It was obvious which cake everyone gathered around, hoping for a taste. Task 1 Task 2
(13)
Tout le monde voulait goûter le gâteau délicieux. % Tout le monde voulait goûter le délicieux gâteau.
[single noun-referent context] Marie planned a birthday party at her house for her friend Christine. She found a recipe for chocolate cake in her new gourmet cookbook. She made the cake, and it looked perfect - but how would it taste? Later that night, Christine blew out the candles on the cake and was offered the first slice. She loved it! She couldn't stop raving about how great it tasted. Now everyone wanted to taste (=goûter) it. Marie hoped she had enough plates. Task 1 Task 2
Tout le monde voulait goûter le délicieux gâteau. Tout le monde voulait goûter le gâteau délicieux.
Given the amount of instruction that learners receive on adjective position in the classroom, it is quite likely that knowledge of explicit rules will be put to use to some degree in accepting or rejecting a particular sentence. For this reason, the following summary of results will concentrate on acceptance rates in the position opposite to the one dictated by instruction. Learners are told that adjectives in Group A (= [1] above), are placed uniquely in pre-N position. Accordingly, we find in Table 1 that they accept them in pre-N position almost equally well under either interpretation (76.2% versus 56%, t = 1.966; p = .063), as indicated by the shaded cells. When these adjectives are placed in post-N position, where we would expect equally high rejection rates, we find instead a non-random distinction based on interpretation: On the basis of "true" responses, there is a significant statistical difference in the acceptability rate between an intersective and nonintersective interpretation, at 63.1% versus 36.9%, respectively (t = 2.624; p = .016).7
6
The % symbol is meant to indicate semantic anomaly. The same results obtain on the basis of "false" answers. In post-N position, the difference in interpretation is significant (t = 2.896; p = .009); in pre-N position it is not (t = 1.603; 7
ADJECTIVE POSITION AND INTERPRETATION
Interpretation pre-N Nonintersective
56%
Intersective
76.2%
37
Acceptance rates post-N 36.9% 63.1%
Table 1: Group A adjectives (pre-N according to instruction) mauvais, vieux, beau, nouveau Unlike the adjectives in Group A, the adjectives in Group (= [2] above) are associated with distinct lexical meanings in the form of English glosses. Learners might therefore be expected to know which meaning is appropriate in which position. As Table 2 shows, there is instead a curious asymmetry: one finds once again a significant difference in the acceptability rate between an intersective and nonintersective interpretation in post-N position at 70.5% versus 43.8% (t = 3.298; p = .004), but no such distinction obtains when the adjective is in pre-N position (56.2%) and 53.3%o, t= .339; p = .738). Interpretation Nonintersective
56.2%
Acceptance rates post-N 43.8%
Intersective
53.3%
70.5%
pre-N
Table 2: Group adjectives (variable according to instruction) ancien, pauvre, cher, grand, propre Turning to the adjectives in Group (= [3] above), which learners are told are uniquely post-N, we find in Table 3 that they accept them in this position in either a single or multiple noun-referent context (78.1%) and 74.3%o, t = .525; p = .605), as indicated by the shaded cells. While learners do accept these adjectives in pre-N position to some extent, there is no difference in acceptability (61.9% and 57.1%, t = .706; p = .489) based on the type of context.
p = .125). Note, however, that even if the distinction in pre-N position were strongly significant, it would be so in the wrong direction: it is a nonintersective interpretation that should be preferred.
38
BRUCE ANDERSON
Multiple noun-referent
61.9%
Acceptance rates post-N 78.1%
Single noun-referent
57.1%
74.3%
Context pre-N
Table 3: Group adjectives (post-N according to instruction) violent, étrange, lourd, délicieux, timide 4.
Discussion The data presented in Tables 1-2 indicate that, aside from an evident bias in acceptance due to explicit rules of adjective placement, intermediate level learners of French demonstrate knowledge of the fact that a nonintersective interpretation in post-N position is significantly less acceptable than an intersective interpretation. The first prediction is therefore supported: results indicate a UG-governed constraint on interpretation to the effect that post-N position (whether as a result of overt noun-movement or some other structural analysis) is to be interpreted intersectively. However, given the availability of both interpretations in pre-N position in Tables 1-3, the second prediction appears not to be supported: a noun-movement analysis of French has not been adopted (at least by the intermediate level). One might point out here that, while the difference in means reported for post-N position in Tables 1 and 2 is statistically significant, it is not particularly robust (26.2% and 26.7%, respectively). However, it is important to keep in mind that there is no way to guarantee that individual participants will react in precisely the same way to each context, nor can one guarantee that the interlanguage grammar in its current state will serve as the only basis for evaluating each token sentence. Nevertheless, the overall within-group pattern of responses for post-N position is non-random and the data split goes in the predicted direction. The lack of a difference in means reported for pre-N position, which is taken to be indicative of a lack of overt noun-movement, must be considered with caution. Recall that the figures in Tables 1-3 rest solely on the ability of the contexts to force one interpretation to the exclusion of the other. For example, the context in (12) presumably forces a multiple noun-referent interpretation to the exclusion of a single noun-referent interpretation. Fearing that contexts such as (12) were in fact too ambiguous between the two interpretations, Anderson (in progress) modified both the context and test sentence as follows:
ADJECTIVE POSITION AND INTERPRETATION
39
(14) [multiple noun-referent context] Monsieur Marat is the judge of a baking contest at the county fair (=foire). He's really looking forward to this contest because there are many old-fashioned cakes made from centuries-old recipes that you just can't taste (=goûter) anywhere else. In tasting all the cakes at the contest, a few turn out to be too dry; others are burnt on the bottom or too bland. Only Madame Dupont's cake turns out to be just right - it practically melts in his mouth. Task 1 Task 2
Monsieur Marat goûte le gâteau délicieux à la foire. % Monsieur Marat goûte le délicieux gâteau à la foire.
In (14), the addition of the sentence-final prepositional phrase à la foire "at the fair" puts the modified noun phrase in sentence-medial position, away from potentially adverse intonation contours. Together with the present tense form of goûter "taste", the PP helps to make the truth-conditional import of the test sentences more apparent than was the case in (12): since Monsieur Marat tastes more than one cake at the fair, the multiple noun-referent reading is less easily defeated, thereby making the pre-N position of délicieux more strongly semantically anomalous. Table 4, from Anderson (in progress), presents the results from 16 learners who were similar to those serving as participants in the present study. The number of adjectives in Group was increased from 5 to 7. Context pre-N Multiple noun-referent
35.7%
Single noun-referent
58%
Acceptance rates post-N 63.4% 87.5%
Table 4: Group adjectives (from Anderson [in progress]) violent, étrange, lourd, délicieux, précieux, épais, charmant Here, the acceptance rates in both post-N and pre-N position are statistically significant based on interpretation. The semantic anomaly of preN adjective position in a single noun-referent context is reflected by a strong difference in means (35.7% versus 58%, t = 4.155; p = .001). A significant difference is now also found in post-N position (63.4% versus 87.5%, t = 5.647; p = .000), demonstrating that learners were sensitive to the change in contextual information. Results such as these would appear to indicate that a noun-movement analysis has been adopted - but if so, why do we not see the reflexes of this parametric change in the acceptance rates of pre-N position for Group A and adjectives (Tables 1-2)? I have no immediate response to this question, short of suggesting similar problems concerning the inability of
40
BRUCE ANDERSON
the contexts to exclude one of the two possible interpretations. Nevertheless, the evidence in Table 4 - which does strongly suggest a parametric change in interlanguage grammars - comes from an interpretive property (viz., the single versus multiple noun-referent distinction) for which there is no relevant instruction or input. It should therefore be considered the most valid indicator of implicit knowledge (i.e., L2 competence). The present study therefore points toward the fruitfulness of investigating UG-constraints on the nominal domain in L2 acquisition beyond mere word order properties, as well as to the need for defining specific poverty of the stimulus problems in research examining patterns for which explicit rules are provided.
REFERENCES Anderson, Bruce. In progress. The Interpretive Correlates of Adjective Position in Native and Nonnative Grammars of French. Ph.D. Dissertation, Indiana University. Bernstein, Judy. 1993. Topics in the Syntax of Nominal Structure across Romance. Ph.D. Dissertation, City University of New York. Birdsong, David. 1992. Ultimate attainment in second language acquisition. Language 68.706-755. Bouldin, John Michael. 1991. The Syntax and Semantics of Postnominal Adjectives in English. Ph.D. Dissertation, University of Minnesota. Chierchia, Gennaro & Sally McConnell-Ginet. 1990. Meaning and Grammar: an introduction to semantics. Cambridge, Mass.: MIT Press. Cinque, Guglielmo. 1994. "On the evidence for partial N movement in the Romance DP". Paths towards Universal Grammar ed. by Guglielmo Cinque et al., 85-110. Washington, D.C.: Georgetown University Press. Coppieters, René. 1987. Competence differences between native and nearnative speakers. Language 63.544-573. Dekydtspotter, Laurent, Rex A. Sprouse & Bruce Anderson. 1997. "The interpretive interface in L2 acquisition: the process-result distinction in English-French interlanguage grammars". Language Acquisition 4.297332. Delbecque, Nicole. 1990. "Word order as a reflection of alternate conceptual construals in French and Spanish: similarities and divergences in adjective position". Cognitive Linguistics 1.349-416. Forsgren, Mats. 1978. La place de l'adjectif épithète en français contemporain. Uppsala, Sweden: Almqvist and Wiksell.
ADJECTIVE POSITION AND INTERPRETATION
41
Gess, Randal & Julia Herschensohn. This volume. "Shifting the DP parameter: a study of anglophone French L2ers". Hawkins, Roger. 1998. "The inaccessibility of formal features of functional categories in second language acquisiton". Paper presented at PacSLRF, held in Tokyo, Japan, March 1998. Heim, Irene & Angelika Kratzer. 1998. Semantics in generative grammar. Oxford: Blackwell. Larsson, Björn. 1993. La place et le sens des adjectifs épithètes de valorisation positive. Lund, Sweden: Lund University Press. Martin, Robert. 1986. "Le vague et la sémantique de l'adjectif. Réflection sur l'adjectif antéposé en français". Quaderni di Semantica 7.243-63. Parodi, Teresa, Bonnie D. Schwartz & Harald Clahsen. 1997. "On the L2 acquisition of the morphosyntax of German nominais". Essex Research Reports in Linguistics 15.1-43. Partee, Barbara. 1987. "Noun phrase interpretation and type-shifting principles". Studies in Discourse Representation Theory and the Theory of Generalized Quantifiers ed. by Jeroen Groenendijk et al., 115-144. Dordrecht: Foris. Partee, Barbara & Mats Rooth. 1983. "Generalized cognition and typeshifting principles". Meaning, Use and the Interpretation of Language ed. by R. Bäuerle et al., 361-383. Berlin: De Gruyter. Schwartz, Bonnie D. & Rex A. Sprouse. 1996. "L2 cognitive states and the Full Transfer/Full Access model". Second Language Research 12.40-72. Valois, Daniel. 1991. The Internal Structure of DP. Ph.D. Dissertation, University of California at Los Angeles. Waugh, Linda. 1977. A Semantic Analysis of Word Order. Leiden: Brill. White, Lydia & Joyce Bruhn de Garavito. 1999. "Against the failed features hypothesis: L2 acquisition of Spanish gender". Paper presented at the 19th Second Language Research Forum, held in Minneapolis, Mn., September 1999. Wilmet, Marc. 1980. "Antéposition et postposition de l'épithète qualificative en français contemporain". Travaux de Linguistique 7.179-204. . 1981. "La place de l'épithète qualificative en français contemporain: étude grammaticale et sylistique". Revue de Linguistique Romane 45.17-73.
FRENCH COMPLEX INVERSION IN THE LIGHT OF A MINIMALIST PROGRAM CEDRIC BOECKX University of Connecticut
1. Pesetsky & Torrego 2000: Displacement = Misplacement Chomsky (2000) suggests that Case and cp-features be looked upon as one and the same entity (two sides of the same coin, really), and proposes that Case be viewed as a signal for attraction, not something that drives movement, but that helps the attractor (probe) identify the feature to attract. Since Case and cpfeatures are lumped together, inherently uninterpretable features no longer exist as they did in Chomsky (1995) (one side of the coin is interpretable on the element, say cp-features on nominais). This is a welcome result, for we now no longer have to address the question of why there are uninterpretable features. More recently, Pesetsky & Torrego (2000) capitalize on the body of work initiated by George & Kornfilt (1981), who first showed that there is an intimate relationship between nominative case and finiteness. It seems overwhelmingly true cross-linguistically that nominative is the Case the element associated with Tense receives. Pesetsky & Torrego give an interesting twist to George & Kornfilt. Whereas Chomsky lumps Case and cp-features together (but nonetheless distinguishes between them), Pesetsky & Torrego argue that Case is really an uninterpretable Tense-feature. By uninterpretable, they actually mean something different from Chomsky (1995). For them, nominative Case is an inherently interpretable Tense-feature that happens to be on the 'wrong' element (a noun, and not a verb), which makes the feature 'uninterpretable.' Pesetsky & Torrego (henceforth P&T) use the idea of Case as uninterpretable T to capture a well-known asymmetry, viz. the need of subject auxiliary inversion with non-subject questions (lb), but its absence from subject questions (la).
CEDRIC BOECKX
44 (1)
a. b.
Who [IP t read the book] What did [IP Mary [VP read t]]
P&T's premises are given in (2) and (3). (2)
(3)
Attract Closest X If a head attracts features a on X, no constituent Y is closer to than X Quantum of closeness: Maximal Projection
Crucially, P&T depart from Chomsky (1995, 2000) in taking Attract Closest to be about categories (X), not features (F). They further rely on a version of Richards's (1997, 1998) Principle of Minimal Compliance (4) to allow movement/attraction from beyond X. Based on (4), they propose that if a movement M crosses more than one Maximal Projection (violating Attract Closest X), there must be a movement M' that crosses only one maximal projection (obeying Attract Closest X). M' renders M licit by Minimal Compliance. (4)
Principle of Minimal Compliance (modeled on Richards 1997, 1998) For any dependency D that obeys constraint C, any elements that are relevant for determining whether D obeys can be ignored for the rest of the derivation for purposes of determining whether any other dependency D' obeys
To apply the above logic to (1), P&T posit an uninterpretable Tense on as the trigger for T-to-C (subject-auxiliary inversion). When T-to-C applies, the way is open for some further movement to reach CP, crossing more than one maximal projection. This accounts for the correlation 'non-subject question'-'T-to-C,' but we now have to explain how the uninterpretable T-feature on is checked in subject questions in the absence of T-to-C. Here the idea of nominative Case as uninterpretable Tense comes into play. If nominative elements have a T-feature, they presumably can check T-features by proxy as it were. This is what P&T claim is taking place in (la). For the account to go through, they make the assumption that features are deleted upon checking, but expunged upon (Phase) Spell-Out. Since IP is not a phase (Chomsky 1999, 2000), the T-feature on
FRENCH COMPLEX INVERSION
45
nominative element is not expunged once it reaches SpecIP. It remains available for checking in the C-domain. Besides the basic asymmetry in (1), P&T are able to reformulate many ECPfacts from the GB-era without positing (any government-based, hence nonminimalist) ECP. The first ECP-facts they account for is the well-known that-t effect in (5). (5)
*who did Mary say that t saw Bill
Again, the starting point is the non-subject close equivalent to (5), (6). (6)
who did Mary say that John saw t
Here movement crosses more than one maximal projection. In particular, in order for the object who to reach the intermediate SpecCP, T-to-C movement must have taken place. P&T argue that this is indeed correct, and that 'embedded' T-to-C in this case can have two forms: either the subject raises to SpecCP, or T raises. In the latter case, the head of the tail is spelled-out as that. P&T claim that the relation between T and that is identical to that found between a clitic and its doubling argument in a language like Spanish (7). The 'doubling' derivation is schematized in (8). (7)
lo vimos a Juan him saw to Juan "We saw Juan."
When that does not surface, as in (9), Nominative-to-C is assumed to take place. The effect is the same: the object can reach SpecCP.
46
CEDRIC BOECKX
The choice between N-to-C or T-to-C is something I will come back to momentarily. For now, it suffices to say that either movement is allowed, N and T being equidistant from C. P&T capture that-t effects straightforwardly (for alleviations of the effect, and a possible way to capture it, see P&T's paper). In case of subject questions, T-to-C is not needed (at least not in the clause where the nominative originates). If P&T are right in taking the presence of that to signal T-to-C, a that-t effect amounts to a violation of Last Resort (unforced T-to-C).
Let us now come back to the choice between T-to-C and N-to-C to license movement from beyond IP. As we saw, the choice seems to be free in embedded contexts like (8)-(9). But clearly, we do want to force T-to-C in matrix questions. Otherwise, something like (11) will be generated.
P&T are aware of the problem. Though they agree that the derivation in (12) should be ruled out, they note that N-to-C is not universally banned from matrix contexts. According to them, (13) is one such case.
The explanation P&T provide is given in (14). (14) "a matrix CP whose head bears [an uninterpretable] Wh[-feature] is interpreted as an exclamative if a non-wh-phrase appears as one of its specifier. Otherwise, it is interpreted as a question"
FRENCH COMPLEX INVERSION
47
P&T thus relate the choice between N-to-C and T-to-C to interpretation. In the remaining of this paper, I examine the well-studied phenomenon of complex inversion in French, and use it as a testing ground for P&T's theory. 2. French Complex Inversion The operation inverting the subject and the verb (/auxiliary) in interrogative contexts is common to many languages of the world. The inversion pattern we find in French (15), however, appears to have peculiar properties. (15) Qui Marie a-t-elle vu Who Marie has-she seen "who did Marie see." Inversion in French is not per se very different from inversion in English: it is restricted to root contexts (16), and it does not take place when the interrogative word is the subject (17). (16) a.
b. (17) a.
b.
*Marie se demande qui Jean aime-t-il Marie SE asks who Jean loves-he "Marie wonders who Jean loves." Marie se demande qui Jean aime Marie SE wonders who Jean loves *Aime qui Marie? loves who Marie "Who loves Marie?" Qui aime Marie? who loves Marie
The most intriguing property of French inversion is the presence of two subjects: Marie and the clitic pronoun elle in (15). Various attempts have been made in the generative literature since Kayne (1972) (see, especially, Kayne 1983, 1986, Déprez 1990, and Rizzi & Roberts 1989), but I think it is fair to say that none of them are satisfactory, especially when looked at from a minimalist point of view: the machinery needed appears as complex as the phenomenon itself. At first, P&T's theory of T-to-C seems to capture many properties of complex inversion. It takes place in the same environment as English T-to-C;
CEDRIC BOECKX
48
French exhibits transparent doubling, either in the form of subject (clitic) doubling, or complementizer doubling in some (mainly Canadian and Belgian) dialects (que in (18)). It also takes place in non-interrogative matrix contexts (19). (18) qui que Marie a vu who that Marie has seen "Who did Marie see?" (19) Quel stupide livre Jean a-t-il lu là What stupid book Jean has-he read there "What a stupid book Jean read!" As a first approximation, one could say that in French (setting aside the dialects that allow (18)) N-to-C applies uniformly. That is, in contexts where English raises T, French raises N. Clitic doubling would be a reflex of N-to-C movement.1 (20) a. b.
[cp Quii [IP ti est [Vp venu]]] [CP Quii Mariej [IP tj a-t-elle [VP vu ti]]]
Since we know that N-to-C is an option even in English, and that it achieves the same result as T-to-C, the proposal appears well motivated. All that needs to be done is parametrize (14). N-to-C in matrix contexts in French would not be tied to non-interrogative interpretation. At this point, questions arise. First, is (14) within the range of possible parameters, which I take to be of the sort argued for in Borer (1984) (see also Chomsky 1995), viz. restricted to "inflectional rules"? Second, how do we explain the position of the subject clitic in (18)? At least since Kayne (1994), enclisis is taken as reflecting V-movement, which crucially is assumed not to take place here. Third, Déprez (1990) and others have provided arguments against subject displacement in French complex inversion. Déprez notes that bare quantifiers like tout "all" cannot be dislocated in French (21), but can nonetheless appear as subjects in a complex inversion context, which shows that it cannot be dislocated.
1
The dialects that allow (18) would have the option of raising T, doubling it by que.
FRENCH COMPLEX INVERSION
49
(21) *Tout, il a été cassé (cf. tout a été cassé) All it has been broken "Everything has been broken." (22) Pourquoi tout a-t-il été cassé? Why all has-it been broken "Why has everything been broken?" In the next section, I address these problems from the perspective of P&T, showing that they cannot receive an adequate treatment. This forces me to offer an alternative. 3. Tackling the problems Besides (14), P&T appeal to another parameter to account for differences in inversion between languages. Thus, they note the following difference pertaining to embedded questions between Standard English and Belfast English. As Henry (1995) shows, Belfast English, in contrast to Standard English, requires inversion in embedded interrogatives. (23) a. b.
She asked who had I seen they wondered what had John done
To account for the difference, P&T suggest that "the dialects differ on one simple point: whether movement [or Agree] is the strategy to delete [the uninterpretable] T[-feature] on Embedded interrogative in Standard English lacks the 'EPP property' [requiring overt movement]." If such a parameter is needed, one wonders whether it could explain the difference between French and English. Could it be that French relies on T-to-C Agree rather than Move? (24) [CP Quii
[IP Marie a-t-elle [VP vu ti]]]
|__Agree_| The latter parameter seems to me to be more plausible. We would indeed be hard-pressed to explain why French complex inversion takes place in both interrogative and exclamative contexts, given (14), which ties the choice between N-to-C and T-to-C to interpretation. By contrast, the analysis in (24) not only removes N-to-C/T-to-C from the realm of interpretation, but provides a possible answer to the other questions raised at the end of the previous section.
50
CEDRIC BOECKX
First, if (24) is right, the subject is not dislocated, and Déprez's criticism is met. Second, and more interestingly in my opinion, (24) allows us to treat the alleged clitic doubling structure as something different. This is welcome, as all the analyses that have tried to account for clitic-doubling in terms of enclisis/spellout of a copy of the subject (trace) have failed to explain why this type of enclisis is restricted to interrogative, or more generally irrealis/emotive/affective contexts. What (24) allows is for us to view the 'clitic' as a case of interrogative (or, more generally mood) inflection, and not as something being related to the subject. Interrogative inflection is well-known from many languages. One thinks here of interrogative particles in languages like Japanese (no). (25) John-wa nani-o katta no? John-topic what-acc bought Q "What did John buy?" More explicitly (from an inflectional point of view), some natural languages have a specific 'interrogative' inflectional morpheme. The oddity of treating what looks like a pronominal element as a case of interrogative inflection dissolves once we bear in mind that, as has been recognized for decades, clitics tend to develop from argumentai to inflectional elements (see Wanner 1987 for the development of Romance clitics). Besides, we know independently that wh-questions affect the inflectional systems in various ways (the phenomenon of 'anti-agreement'; the disruption of subjectverb agreement in various dialects of English (Kayne 1989), etc.) The very existence of cases like Palauan actually led Rizzi (1996) to claim that the Q-feature checked in C° starts off on T/I° (see also Hagstrom's 1998 claim that the Q-marker in languages like Japanese does not start in C, but inside the clause). Furthermore, it has long been recognized that interrogation relates to focus, and the literature is replete with works showing the connection between focus and inflection (see Kiss 1995 for a comprehensive overview). So, upon closer scrutiny, the idea of treating the clitic il in (24) as interrogative inflection appears not so ill-motivated. Note that it immediately dissolves the two-subject problem in inversion contexts. If one of the subjects is not a true subject (or argumentai morpheme), but an interrogative marker, there is only one subject in the clause. Furthermore, it allows us to at least suggest the absence of overt T-to-C in French. Cheng
FRENCH COMPLEX INVERSION
51
(1991) proposes as a universal parameter (her Clause-Typing hypothesis) that the presence of a Q-marker (for our purposes, read, interrogative inflection) correlates with the absence of movement. For her, movement means whmovement (this is how she captures the wh-fronting/wh-in-situ divide), but I would like to extend it to the case at hand, and claim that if a language has interrogative inflection, it lacks overt T-to-C (i.e., Agree suffices).2 As to the question of why French uses an otherwise pronominal marker for interrogative inflection, it could be a morphological accident, but I believe that something deeper is at play. There is an intricate relationship between Case and Agreement morphology (see Chomsky 2000 and the references there). Likewise, we know of the relationship between Case and Tense (P&T), between Tense and Focus (Motapanyane 1998), between Case and Focus (Schütze 1999), between Focus and questions (Horvath 1986), and even between agreement of focus (Simpson and Wu 2000). Putting all of that together, one arrives at some plausible (though admittedly not fully understood) relationship between agreement and interrogation (or mood, more generally), which is precisely what we find in French, which uses an agreement marker (clitic) as interrogative inflection. As noted by Karlos Arregui (p.c.), French does not collapse agreement and interrogative inflection (the verb in T still agrees with the subject in (28)); it seems to make use of some reduplicative process (where the reduplicant takes the form of the clitic) ~ a morphological process which I suspect should be easily formalizable in a framework like Distributed Morphology, on a par with cases of fission, and elsewhere exponent (Halle & Marantz 1993, Halle 1997). In sum, treating the 'subject-clitic' as a(n interrogative) mood marker of sort appears well-motivated cross-linguistically, and allows me to account for the basic properties of French complex inversion in a way consistent with otherwise well-motivated morphological/inflectional parameters. To sum up the analysis in one line: French does not display overt T-to-C, but mere Agree; 'replacing' displacement by inflection.3 2
This might be too strong. It could be that some languages exhibit T-to-C movement independently of question formation (some VSO, or Verb-second languages, for instances). Such cases, if they exist, should be kept separate from French-type languages. The interrogative would act as a free rider in those cases. 3 The analysis proposed here bears some similarity with Sportiche's (1998) analysis of complex inversion, where it is also claimed that French lacks overt T-to-C. Sportiche's framework, however, is sufficiently different from the one I assume here for me to refrain from attempting a comparison between the two analyses. Abbas Benmamoun (p.c.) informs me that he also suggested in unpublished work treating the doubling clitic in French as some Q-marker.
CEDRIC BOECKX
52
If I am correct, there is no N-to-C in French (at least, this is not needed), as Agree between T-to-C counts as a valid case of Attract Closest X (here, X = F(eature)) to open the gate for further movement.4 A last point I would like to address is the existence of matrix questions with overt complementizers, as in (18) (repeated here as (26)). (26) qui que Marie a vu? I would like to analyze such cases the same way P&T treat that, viz., as a reflex of T-to-C. Here the complementizer que is doubling T. It is quite plausible, I think, to assume that complementizer doubling of this kind will replace the less transparent case of T-to-C in the form of clitic doubling, as the latter introduces morphological ambiguity between 'pure' agreement and mood marker. It is not a surprise from this point of view to find that (26) is found in the most 'innovative' ('modern') dialects of French. Finally, let me mention the existence of interrogative sentences like (27) in some dialects, where neither clitic-doubling nor complementizer-doubling has taken place. To avoid positing N-to-C in this case (or some 'large-scale' parameter), I take it that the interrogative inflection is not realized morphologically in this case (T-to-C Agree has no reflex here). (27) qui Marie a vu? Who Marie has seen "Who did Marie see?"
We saw on the basis of data discussed by Déprez (1990) that the absence of N-to-C is wellmotivated. Yet, many speakers feel that the subject is dislocated in complex inversion (as compared to regular declarative sentences, where the subject appears clause-initially). It is possible that the sensation of dislocation comes from a more articulated structure in (25). It would indeed be interesting to pursue the idea that the presence of interrogative inflection counts as sufficiently rich inflection to license pro-drop in French (an otherwise non-pro-drop language) - making the overt subject dislocated. Pollock (1998) makes a good case for (restricted) pro-drop and 'dislocated' subjects related to clitics in structures involving en. The structure he provides is something like (i). (i) [T [subj Jean pro] [T en(v-)Vφ ]] Adopting such a structure for (28) would yield something like (ii). (ii) [T [subj Jean pro] [ V-ν- φ-Q ]] I leave it for future research to determine whether (ii) is a viable structure for French, as it does not directly affect the argument made here.
FRENCH COMPLEX INVERSION
53
This should not come as a surprise, as Cheng (1991) already noted that question particles often go unpronounced in languages that have them. This seems to be the case in French as well, as many dialects that allow (27) also allow the more formal version with clitic doubling (some dialects, mine included, even allow the three possibilities: zero-inflection, clitic doubling, and complementizer doubling).5 4. Conclusion: Pesetsky & Torrego (2000) Revisited We have seen that French complex inversion poses serious problems if analyzed along the lines of P&T's account of T-to-C in English. We further saw that internal to P&T's account, problems of learnability (the nature of parameters) appear. This begs the question of whether P&T's account should be maintained for English. I believe that their conception of Case as uninterpretable ('misplaced') Tense is fundamentally right (Alexiadou & Anagnostopoulou 1999, and Boeckx 2000, forthcoming, among others, provide evidence that Chomsky's 2000 equation Case = Phi-feature is not tenable). By contrast, their motivation for T-to-C (uninterpretable T -feature on C) runs counter to recent attempts (Chomsky 1995, 1999, 2000; Boeckx & Stjepanovic forthcoming) to eliminate headmovement (as feature-driven) altogether. In addition, it will be recalled that P&T's premise is Richards's Principle of Minimal Compliance. The nature of the phenomena discussed in Richards (1998) suggest that we are not dealing with a natural class; and the prime motivation for Minimal Compliance in Richards (1997) (multiple wh-fronting) has been given a superior treatment in Boškovié (1999, 2000)6. This makes me think that if the non-standard version of Attract Closest X at the heart of P&T is not adopted, the whole analysis dissolves. The question then is: Is there an alternative? I believe there is. Based on the distribution of Quantifier-Float in Irish English, McCloskey (2000) argues that the EPP-requirement on SpecTP is No dialect, as far as I am aware, allows (i). qui que Marie a-t-elle vu t? who that Marie has-she seen "Who did Marie see?" This gap is expected under our approach, which takes que-complementizer and inversion to be two reflexes of the same process (T-to-C). 6 Even Richards (1999) has shown that the logic of Minimal Compliance leads one to a model of grammar that is radically different, and possibly incompatible with P&T's analysis. (i)
54
CEDRIC BOECKX
shifted to SpecCP in subject questions (see McCloskey 2000 and Boeckx forthcoming for various ways of capturing this 'lifting' process). As noted in Boeckx (forthcoming), this immediately removes the 'vacuous movement' problem (do subject wh-phrases move to SpecCP or do they stay in SpecIP): Subject wh-phrases never undergo vacuous movement, as they never occupy SpecIP. Further, it immediately explains why no Jo-support is found in subject questions, even if CP is targeted: nothing, not even a subject wh-trace intervenes between and T/V. Affix-Hopping (or whatever accounts for the distribution of the verb) can proceed unhindered. By contrast, in non-subject questions, the subject blocks affix-hopping (see Watanabe 1993 and Boskovic 1998 for evidence that C° is affixal in English), triggering do-support. This is essentially what P&T wanted to account for. Note that nothing like Attract Closest X, Minimal Compliance, or feature-driven head-movement is needed. Nor do I need to appeal to the ECP (a virtue of P&T's analysis), which suggests that the above sketch, once fully fleshed out, might provide a viable alternative to P&T. The other phenomena P&T discuss (ECP-extensions like that-t effects) would then have to be treated separately (for a possible analysis, see Boeckx forthcoming). Be that as it may, whether P&T's analysis can be dispensed with or not, I have shown that French complex inversion receives an adequate treatment if one is ready to adopt the claim that the most salient characteristic of the construction (subject clitic doubling) is a case of interrogative inflection, which allows T to relate to without actual movement.
REFERENCES Alexiadou, Artemis & Elena Anagnostopoulou. 1999. "Raising without and the nature of agreement". In Proceedings of WCCFL 18 ed. by S. Bird, A. Carnie, J. Haugen & P. Norquest, 14-26. Somerville, Mass.: Cascadilla Press. Boeckx, Cedric. 2000. "EPP eliminated". Ms., University of Connecticut. . forthcoming. Restrictions on Syntactic Patterns. Ph.D. Dissertation, University of Connecticut. & Sandra Stjepanovic. Forthcoming. Head-ing toward PF". Linguistic Inquiry 32. Borer, Hagit. 1984. Parametric Syntax. Dordrecht: Foris.
FRENCH COMPLEX INVERSION
55
Boskovic, Zeljko. 1998. "LF-movement in the minimalist program". Proceedings of NELS 29 .43 -57. University of Massachusetts, Amherst: GLSA. . 1999. "On multiple feature-checking". Working Minimalism ed. by S. D. Epstein & N. Hornstein, 159-187. Cambridge, Mass.: MIT Press. . 2000. "What is special about multiple wh-fronting". Ms., University of Connecticut. Cheng, Lisa. 1991. The Typology of Wh-questions. Ph.D. Dissertation. MIT. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. . 1999. "Derivation by phase". MITOPL #17. To appear in Ken Hale: A Life in Language ed. by M. Kenstowicz. Cambridge, Mass.: MIT Press. . 2000. "Minimalist inquiries: the framework". Step by Step: essays in minimalism in honor of Howard Lasnik ed. by R. Martin, D. Michaels, & J. Uriagereka, 89-155. Cambridge, Mass.: MIT Press. Déprez, Viviane. 1990. "Two ways of moving the verb in French". MIT Working Papers in Linguistics #13, Papers on Wh-movement ed. by L. Cheng, & H. Demirdache, 47-85. Cambridge, Mass.: MITWPL. George, Leland M. & Jaklin Kornfilt. 1981. "Finiteness and boundedness in Turkish". Binding and Filtering ed. by Frank Heny, 105-127. Cambridge, Mass.: MIT Press. Hagstrom, Paul. 1998. Decomposing questions. Ph.D. Dissertation, MIT. Halle, Morris. 1997. "Impoverishment and fission". MIT Working Papers in Linguistics #30. PF: Papers at the Interface ed. by B. Bruening, Y. Kang & M. McGinnis, 425-450. Cambridge, Mass.: MITWPL. & Alec Marantz. 1993. "Distributed morphology and the pieces of inflection". In The View from Building 20, ed. by K. Hale & S. J. Keyser, 111176. Cambridge, Mass.: MIT Press. Henry, Alison, 1995. Belfast English and Standard English. Oxford: Oxford University Press. Horvath, Julia. 1986. Focus In the Theory of Syntax and the Syntax of Hungarian. Dordrecht: Foris. Kayne, Richard. 1972. "Subject inversion in French interrogatives". Generative Studies in Romance Languages ed. by J. Casagrande & B. Sciuk. Rowley, Mass.: Newbury House. . 1983. "Chains, categories external to S, and French complex inversion". Natural Language and Linguistic Theory 1.107-139. . 1986. "Connexité et inversion du sujet". La Grammaire Modulaire ed. by M. Ronat & D. Couquaux. Paris: Editions de Minuit.
56
CEDRIC BOECKX
. 1989. "Notes on English Agreement". Ms., CUNY. . 1994. The Antisymmetry of Syntax. Cambridge, Mass.: MIT Press. Kiss, Katalin E., ed. 1995. Discourse-configurational Languages. Oxford: Oxford University Press. McCloskey, James. 2000. "Quantifier float and wh-movement in an Irish English". Linguistic Inquiry 31.57-84. Motapanyane, Virginia. 1998. "Focus, checking theory, and fronting strategies in Rumanian". Studia Linguistica 52.227-243. Pesetsky, David & Esther Torrego. 2000. "T-to-C Movement: Causes and Consequences". To appear in Ken Hale: A Life in Language ed. by M. Kenstowicz. Cambridge, Mass.: MIT Press. Pollock, Jean-Yves. 1998. "On the syntax of subnominal clitics: cliticization and ellipsis". Syntax 1.300-330. Richards, Norvin. 1997. What Moves Where When in Which Language. Ph.D. Dissertation, MIT. . 1998. "The Principle of Minimal Compliance". Linguistic Inquiry 29.599-629. . 1999. "Dependency formation and directionality of tree construction". MIT Working Papers in Linguistics #34: Papers on Morphology and Syntax ed. by V. Lin, C. Krause, . Bruening & . Arregi, 69-105. Cambridge, Mass.: MITWPL. Rizzi, Luigi. 1996. "Residual Verb-Second and the wh-criterion". Parameters and Functional Heads ed. by A. Belletti & L. Rizzi. Oxford: Oxford University Press. & Ian Roberts. 1989. "Complex inversion in French". Probus 1.1-30. Schütze, Carson. 1999. "Case stacking in Korean". Ms., UCLA. Simpson, Andrew, and X-Z Z. Wu. 2000. "Agreement, shells, and focus". Presented at WCCFL 2000. Sportiche, Dominique. 1998. Atoms and particles of clause structure. London: Routledge. Wanner, Dieter. 1987. The Development of Romance Clitic Pronouns from Latin to Old Romance. Berlin: De Gruyter. Watanabe, Akira. 1993. AGR-based Case Theory and Its Interaction With the A-bar System. Ph.D. Dissertation, MIT.
OPTIONALITY, PRESUPPOSITION, AND WH-IN SITU IN FRENCH CEDRIC BOECKX, PENKA STATEVA & ARTHUR STEPANOV University of Connecticut
0.
Introduction It is well-known that in matrix contexts French combines English and Chinese properties when it comes to wh-questions. Like English, French fronts wh-phrases; but it also allows them to stay in situ, much like Chinese. This is demonstrated in (1), (2) and (3): (1) (2)
a. b. a.
(3)
b. a.
b.
What did John buy? *John bought what? *Sheme mai-le John? what buy John "what did John buy?" John mai-le sheme? Qu'a acheté John? what-has bought John "what did John buy?" John a acheté quoi?
English Chinese
French
The optionality that French exhibits is problematic on several grounds. For one thing, it casts doubt on the "Chengian" view on the typology of whquestions (Cheng 1991, Tsai 1994, Hagstrom 1998), which seeks to motivate the cross-linguistic variation found in wh-strategies by capitalizing on the nature of (simplifying dramatically) wh-words and question-particles. For another, French seems to flatly violate the 'minimalist' ban on 'pure' optionality (optionality of morphologically-driven movement), fronting appearing at first not to be a Last Resort option in (3b).1
1 Cheng and Rooryck (forthcoming) propose an account of w/z-in-situ in French which is compatible with Cheng's (1991) Clause-typing hypothesis, but far from compatible with minimalist assumptions, as has been shown in Boeckx (2000).
58
CEDRIC BOECKX, PENKA STATEVA & ARTHUR STEPANOV
There have been some recent approaches trying to make sense out of the French paradigm just given. We will here concentrate on the most detailed, and empirically adequate of these: Boskovic (1998, 2000). Boskovic (1998) (see also Boskovic 2000, Boskovic & Lasnik 1999) capitalizes on the possibility left open in Chomsky's (1995) system for covert insertion of phonologically null elements, and argues that French C° is such an element. If it enters the structure prior to Spell-Out (i.e., insertion in the 'overt' component), wh-fronting is overt, as in (4a). If C° is inserted covertly, wh-movement takes place covertly, as in (4b). (According to Boskovic, has a strong feature which triggers wh-movement whenever it is inserted.)
Boskovic's analysis is designed to capture the root vs. long-distance asymmetry concerning wh-in situ: He claims that French disallows wh-words in situ in long-distance contexts, which sets French apart from languages like Japanese or Chinese, as shown in (5): (5)
a.
b. cf. c.
Qu' a dit Pierre que Jean a acheté t? What has said Pierre that Jean has bought "What did Pierre say that Jean bought?" *Pierre a dit que Jean a acheté quoi? (* as non-echo) Pierre has said that Jean has bought what Lisi shuo [shei mai-le shu]? Lisi says [who buy-asp books] "What does Lisi say bought books?"
Chinese
According to Boskovic, If is inserted at LF, movement is restricted to formal features of the wh-phrase only. Boskovic argues, following Chomsky (1995), that feature-movement amounts to head-movement. This provides a nice explanation for why LF-movement seems more restricted than overt movement of wh-phrases in French. If feature-movement is head-movement, we expect heads to matter when it comes to locality. We know independently that head-movement is much more severely restricted than XP-movement. In the particular case of French, Boskovic claims that an intermediate C-head along the wh-path intervenes, as schematized in (6). This straightforwardly accounts for the impossibility of wh-in-situ in long-distance contexts.
OPTIONALITY. PRESUPPOSIION, AND WH-IN SITU IN FRENCH
(6)
[C° Pierre a dit [ que Jean a acheté quoi]
59
(=5b)
Boskovic also discusses embedded questions in French. He claims that wh-in situ are not allowed in embedded clauses, as shown in (7a) they are bad, in contrast with the embedded wh-movement case in (7b): (7)
a.
b.
*Pierre a demandé tu as vu qui Peter has asked you have seen who "Peter asked who you saw." Pierre a demandé qui tu as vu Peter has asked who you have seen
Under Boskovic's approach, wh-in situ are disallowed in embedded contexts due to the severe restrictions on covert insertion of lexical items. Basically, insertion is only allowed to take place at the root of the tree. Furthermore, LF-insertion is only restricted to phonetically-null elements. Given the necessary, phonologically non-null character of an intermediate projection in embedded tensed contexts, the intermediate C° cannot be inserted at LF (merger would not take place at the root of the tree). In this paper, we argue against several aspects of Boskovic's characterization of wh-in-situ in French. First, it is possible for wh-phrases to remain in situ in long-distance contexts, when the matrix verb is non-intensional, (e.g. factive) as in (8).2 (8)
Jean sait/ regrette/a découvert que Marie a acheté quoi? Jean {knows, regrets, found out} that Marie has bought what "What does Jean {know, regret, find out} that Marie bought?"
This constrasts with questions with intensional matrix verbs, shown in (9): (9)
*Jean pense/ croit que Marie a acheté quoi? Jean {thinks, believes} that Marie has bought what "What does Jean {think, believe} that Marie bought?"
Such contrast is unexpected under Boskovic's analysis. 2
It is important to realize that not all dialects of French exhibit the contrasts discussed here and below. What follows is thus appropriately seen as a characterization of French wh-in-situ, as it is used in some varieties of Belgian French. The cross-dialectal differences are addressed in Boeckx (2000)
60
CEDRIC BOECKX, PENKA STATEVA & ARTHUR STEPANOV
The second objection concerns embedded questions. Examples like (7a) are indeed not acceptable, as Boskovic shows. However, there is an interfering factor, namely, French disallows null complementizers. Once we plug in a complementizer, (7a) becomes good, which is shown in (10): (10) Pierre a demandé si tu as vu qui Peter has asked whether you have seen who "Peter asked who you saw." According to our judgments and those of our informants (bearing in mind the caveat in note 2), embedded questions are fine with wh-in-situ irrespective of the nature of the embedding verb. This is shown in (11) and (12), respectively. (11) Pierre se demande si Marie aime qui Pierre wonders whether Marie loves who "Pierre wonders who Marie loves." (12) Pierre a oublié que Marie aime qui Pierre has forgotten that Marie loves who "Pierre forgot who Marie loves." Still within the realm of embedded questions, it is worth mentioning that a wh-phrase can also be left in situ in a clause embedded under expressions like "Tell me", as shown in (13): (13) Dis moi: Jean a acheté quoi Tell me Jean has bought what "Tell me what Jean has bought." In order to account for the distribution of w/z-in-situ in French we just reviewed, it is crucial to go into the details of its semantics, which we do in the next section. 1. Interpretive differences While it is felicitous to answer a question like (14a) (where fronting has taken place) by 'nothing' in French, it is not felicitous to answer a question like (15a) (where the wh stays in situ) by 'nothing.' (Infelicity is indicated by the diacritic '*!'.) This fact was, to our knowledge, first noted in Chang (1997).
OPTIONALITY, PRESUPPOSIION, AND WH-IN SITU IN FRENCH
(14) a. b. (15) a.
b.
61
Qu' a acheté Jean? "What has Jean bought?" Un livre/une voiture/rien a book/a car/ nothing Jean a acheté quoi? Jean has bought what? "What has Jean bought?" Un livre/une voiture/*!rien
To further clarify the difference in the usage of wh-in-situ and fronted wh-phrases, consider the following situation. Suppose all discourse participants share the information that John went shopping. Even in this case it is still not appropriate to open an exchange with the question (15a). However, that opening of a conversation becomes acceptable if the discourse participants know that Jean bought something when he went shopping. What this example shows is that the wh-in-situ requires a familiarity of a particular type. Using Heim's (1982) terminology, we can say that the context has necessarily been changed from its initial state to a state with added information/a card for an item bought by John in order for the question with the wh-in-situ to be felicitously uttered. On the other hand, the question in (14a) with a fronted wh-phrase is felicitous as an opening sentence in a discourse where the participants share no background information before that utterance. The fronted wh-phrase requires no familiarity. In Heim's framework, familiarity is related to defmiteness. Simplifying her ideas, we conclude that the wh-in-situ in (15a) has to be treated on a par with definite NPs. These facts strongly indicate that the crucial semantic factor that plays a role in the contrast between questions with wh-movement and wh-in-situ in French is presuppositionality. Note that the phenomenon of wh-in-situ in French should not be equated with "echo-questions," which also involve presupposition, but exemplify a different phenomenon altogether. In the case of echo-questions, as standardly understood, discourse participants not only know that Jean bought something, but the answer is already part of discourse. The only thing about echoquestions is that the answer is not accepted into the common ground by the speaker (due to surprise, failure to hear, etc.). By contrast, in the case of whin-situ in French, the speaker knows that Jean bought something, but he does not know what.
62
CEDRIC BOECKX, PENKA STATEVA & ARTHUR STEPANOV
In order to understand the role of presupposition for French wh-in-situ better, a little excursus into interrogative strategies in French will prove useful. French has four strategies to form questions, as shown in (16): fronting, in-situ, reinforced fronting and cleft: (16) a.
b.
d.
Qui as-tu vu? who have-you seen "Who did you see?" Tu as vu qui? you have seen who? "who did you see?" Qui est-ce que tu as vu? who is it that you have seen "Who is it that you saw?" (C'est) qui que tu as vu? (it is) who that you have seen "Who is it that you saw?"
Fronting
In-situ
Reinforced fronting
Cleft
When we concentrate on 'possible felicitous answers,' we see that fronting and reinforced fronting pattern alike in allowing an answer like 'nothing.' By contrast, in-situ and cleft strategies disallow such an answer. There is further interesting evidence that in-situ and cleft strategies pattern alike. Some wh-words in French have both strong and weak (cliticlike) forms (in roughly the sense of Cardinaletti & Starke 1999). For instance, the object wh-element que ('what') can surface as qu' or quoi, thus resembling non-wh pronouns like 3rd person le, which has a weak and a strong form, and ça, respectively. This is shown in (17). Depending on the syntactic and prosodic contexts, only one form is allowed to surface. (17) a.
b.
Qu' as-tu vu? What have-you seen "What did you see?" Qu'est-ce que tu as vu? What-is-it that you have seen "What is it that you saw"? Tu as vu quoi? You have seen what "What did you see?"
Fronting
Reinforced fronting
In-situ
OPTIONALITY, PRESUPPOSIION, AND WH-IN SITU IN FRENCH
d.
(C'est) quoi que tu as vu? It is what that you have seen "What is it that you saw?"
63
Cleft
Given that the distribution of the strong and weak forms appears to be governed by the amount of focus put on the element, let us examine whether focus might explain some of the restrictions we saw on felicitous answers. It seems that it does. Thus, much like it is impossible to answer (17c) by 'nothing,' it is impossible to use an element like nothing in a cleft structure, as shown in (18). (18) *!It is nothing/nobody that John saw Note in this respect that English disallows an answer like 'nothing' to a cleft-question like (19). (19) a. b.
What is it that John bought? *!Nothing
Also, we know that NPIs cannot be clefted, as shown in (20). (20) *It is ANYTHING that John did not buy Likewise, the French NPI je ne sais quoi ('I-don't-know-what') is disallowed in an in situ context, as shown in (21). (21) *Il a vu je ne sais quoi? He has seen I-don't-know-what "*He saw anything" Given that, we would like to propose that in situ wh-phrases in French are focused, which is why they cannot be felicitously answered by 'nothing.' However, as we will show immediately, we cannot analyze French wh-in situ and clefts in a uniform manner. 2. The 'empty' D proposal Although the analogy between French w/z-in-situ and cleft strategies was very helpful in highlighting the role of presupposition, one should be aware of differences between the two strategies, which prevents us from giving a unified account to both strategies. First, as has been often noted by various
64
CEDRIC BOECKX, PENKA STATEVA & ARTHUR STEPANOV
authors, including Boskovic, negation prevents the in-situ strategy in matrix contexts. This is shown in (22): (22) *Jean n 'a pas acheté quoi Jean not-has not bought what "What didn't John buy?" However, negation does not cancel the interrogative reading of the whphrase in French clefts, as (23) demonstrates: (23) C'est quoi que Jean n'a pas acheté? It is what that Jean not-has not bought "What it is that Jean did not buy?" So, wh-in situ in French is different from the cleft construction. We have also established that the interpretive differences between questions involving wh-in situ and moved wh-in French amount to different behavior of presupposition in these two types of questions. The presupposition plays a crucial role in the wh-'m situ version, but not in the moved wh-version. It is natural, therefore, to view this contrast with respect to presupposition as a consequence of corresponding structural difference(s). The direction we would like to pursue is that presuppositionality is reflected in the morphological make-up of wh-phrases in situ. Specifically, we propose that the wh-phrase in situ (focused) is headed by a phonologically silent definite head D. The D head is responsible for the existence presupposition, and also for the uniqueness implication in questions with wh-'m situ. The internal structure of a wh-phrase in situ would be something like (24):3 (24) [DPD [wh-phrase]] Naturally, the D head is not present in the wh-phrase when the latter undergoes movement. We further assume, slightly departing from Chomsky (1995), that the Q feature of the interrogative complementizer can be checked either by the wh-feature, or, alternatively, by the D feature of the wh-phrase. 3
Cristina Schmitt (personal communication) notes that Brazilian Portuguese, which has optional whfronting, uses an overt determiner when the equivalent to what stays in situ (o que instead of que), thus offering striking confirmation for our hypothesis. She also notes that wh-in-situ is associated with the same kind of presupposition that we describe here for French. Unfortunately, the parallelism between French and Portuguese is not perfect, as Cheng and Rooryck (forthcoming) have noted. We leave for future research the investigation of the differences between the two languages (see Boeckx 2000 for preliminary remarks).
OPTIONALITY, PRESUPPOSIION, AND WH-IN SITU IN FRENCH
65
We will now see how our proposal derives the observed interpretational contrast. Let us look again at the baseline French examples repeated in (25): (25) a.
b.
Qu'a a acheté Jean? what has bought Jean "What did John buy?" Jean a achete quoi? Jean has bought what
In (25a) the bare wh-phrase checks the Q feature of the interrogative complementizer. In (25b), the Q feature is checked, rather, by the silent D head. Movement of D is thus an instance of A'-head movement in the sense of Roberts (1992). The wh-phrase itself remains in situ and the D feature, now in C, induces the presuppositional context of (25b). Similar considerations apply in the case of the French embedded questions. The only difference is that the interrogative complementizer is introduced by verbs like wonder, as shown in (26): (26) a. b.
Jean se demande qui Pierre aime Jean wonders who Peter loves Jean se demande si Pierre aime qui Jean wonders if Pierre loves who
Let us now examine the long-distance cases. We observed the following regularity: if the wh-phrase is embedded under a matrix intensional verb of the think-type, it must move to the matrix C°; it cannot stay in situ. This is exemplified in (27). If, on the other hand, the wh-expression is embedded under a non-intensional verb, the wh-phrase can be fronted or left in-situ as in (28): (27) a.
b. (28) a.
b.
Qui Jean pense-t-il que Pierre aime? Who Jean thinks-he that Pierre loves "Who does Jean think that Pierre loves?" *Jean pense/croit que Pierre aime qui Jean thinks/believes that Pierre loves who Qui Jean sait-il/regrette-il que Pierre aime Who Jean knows-he/regrets-he that Pierre loves "Who does Jean know/regret that Pierre loves?" Jean regrette/sait que Pierre aime qui Jean regrets/knows that Pierre loves who
66
CEDRIC BOECKX, PENKA STATEVA & ARTHUR STEPANOV
The paradigm shows that wh-extraction is sensitive to the semantic properties of the matrix verb, therefore a semantic account for the facts is in order. Consider first how the possible LF for each of the above sentences is derived. According to our proposal, the wh-phrases in the (a) examples differ from the ones in the (b) examples in that the latter are focused and have a different syntactic structure: they are headed by a silent D°. (27a) and (28a) are derived by overt fronting of the wh-phrase which checks the Q-feature of the complementizer. If the wh-phrase is focused, as in (27b) and (28b), the wh-word cannot move overtly because of the Specificity constraint (cf. Chomsky 1973) which bans extraction out of specific NPs. As discussed earlier, D° can potentially check the Q-feature of C° so there is a possibility to adjoin D° to C°. Yet, in order for us to arrive at an interpretation for the sentence of the appropriate type, we need to move the complement of D°, the wh-phrase, close to it at LF. However, if the matrix verb is intensional, scoping a definite description above it would change the presuppositions of the sentence. In order to explain how this happens, we will first discuss an example with a simple definite NP and then we will return to our case of a definite wh-phrase. Karttunen (1974) and Heim (1992) observed that intensional verbs 'filter' the presuppositions of their complements in a specific way. Consider for example (29): (29) John thought that he talked to the son of Bill Clinton. The definite article presupposes the existence of its complement and the embedded clause is assigned a meaning only if the context satisfies the presupposition that Bill Clinton has a son. Still, the whole sentence does not inherit the presupposition of the embedded clause. The person who utters (29) is not committed to the belief that Bill Clinton has a son. For (29), however, to be assigned an interpretation it is necessary that the discourse participants share the presupposition that John believes that Bill Clinton has a son. So intensional verbs then alter the presuppositions of their complements in a very particular way: a presupposition about the existence of what is denoted by the sister of the definite article percolates up to the matrix clause as a presupposition about the beliefs that the subject of the intensional verb holds. Now, we are ready to go back to (27b) and (28b). Nothing would go wrong syntactically if D° checks the wh-feature of C° in both examples. However, if at LF D° is structurally higher than the matrix verb, then a
OPTIONALITY, PRESUPPOSIION, AND WH-IN SITU IN FRENCH
67
context that could satisfy the presuppositions of both sentences should be one in which the discourse participants believe that Pierre loves someone. In (28b) where we have a non-intensional verb the presupposition that the focused wh-phrase introduces is free to project to the matrix level. (27b), on the other hand, is unacceptable for that very reason: by scoping D° above the intensional verb, we let the presupposition associated with it to project to the matrix clause. However, the native-speaker's judgment in this case is that what is presupposed in (27b) is the belief of the subject, Jean, that Pierre loves someone, not the belief of the discourse participants. (27b) then is ruled out for semantic reasons, namely we get wrong presuppositions for it if we raise D° above the intensional verb. 3. Independent evidence: Lahiri (1991) An independent argument supporting our proposal comes from French embedded questions. The data in (30) shows that 'know' can take a sentential complement in which the wh-phrase can either move or stay in situ. (30) a. b.
Jean sait qui Pierre a embrassé "Jean knows who Pierre has kissed." Jean sait que Pierre a embrassé qui Jean knows that Pierre has kissed who "Jean knows who Pierre kissed."
However, in (31) the wh-phrase cannot stay in situ, which is predicted by our analysis. (31) a.
b.
Jean sait où acheter de l' essence pour sa voiture Jean knows where buy some gas for his car "Jean knows where to buy gas for his car." *Jean sait acheter de l'essence où pour sa voiture "Jean knows to buy gas for his car where."
It has been standardly assumed since Karttunen (1974) that the knowclass verbs take complements with exhaustive interpretation. The examples in (30) illustrate this requirement of the verb. We interpret both sentences to have a presupposition that Pierre kissed someone (since know is a factive verb) and to assert that for every person that Pierre kissed, Jean knows that Pierre kissed that person. (30a) and (30b) end up with the same interpretation although semantically and syntactically they are derived in a different way. To put it in terms of our proposal, in (30b), an empty D-head takes the wh-
68
CEDRIC BOECKX, PENKA STATEVA & ARTHUR STEPANOV
phrase as a complement. D° presupposes the existence of the content of its complement (wh-phrase) and implies its uniqueness. The implication is translatable into the requirement for exhaustivity of the set denoted by the wh-phrase. Because of D°, (30b) necessarily means that Jean has knowledge about each person being kissed by Pierre. (30a) differs from (30b) because it has a 'bare' wh-phrase. Still, it gets the same presupposition by virtue of having the factive verb in the matrix clause. In addition, we understand the complement of know to denote the maximal set of true sentences of the type 'Pierre kissed x'. (30b) is slightly dispreferred. On the one hand, we showed that D° presupposes the content of his complement and the embedded clause inherits that presupposition and the uniqueness implication. On the other hand, the same presupposition and implication can be attributted to the role of the factive verb. We suppose that it is this redundancy that has some negative effect on acceptability. Turning now to (31), we notice that we do not understand (31a) as requiring an exhaustive answer. This fact was observed in Lahiri (1991). He showed that there are cases which constitute an exception for the exhaustivity requirement (for further discussion, see Beck and Rullmann 1999). The observation is that for a speaker to truthfully claim (31a) it suffices if Jean knows at least one place where he could buy gas for his car (this would satisfy the presupposition requirement). No doubt, the speaker does not have to refer to the maximal set containing all the places where Jean could potentially buy gas. On the other hand, if our theory about the morphological make-up of the focused wh-phrase is correct, we would predict that in (31b), the speaker would need to make a claim about the maximal set of possible referents for where. Since this is pragmatically implausible in this example, we expect (31b) to be unacceptable. The prediction is borne out. We have seen that our present analysis captures the entire paradigm involving French wh-in situ. We correctly predict that matrix wh-questions should pattern with embedded wh-questions in French in allowing both the in-situ version and the wh-movement version. We also predict, in accord with the facts, that long-distance questions with wh-in situ are not possible in French if the matrix verb is intensional {think, believe) and are possible if the matrix verb is not intensional, in particular, factive {know). Our account is superior to the analysis of the similar paradigm in Boskovic (1998, 2000) in a number of important respects. First, and most straightforwardly, it takes into account the interpretive differences between the moved and the in situ versions with respect to presuppositionality.
OPTIONALITY, PRESUPPOSIION, AND WH-IN SITU IN FRENCH
69
Second, recall that Boskovic argues that in embedded questions the interrogative complementizer has to be inserted in overt syntax and trigger wh-movement to check the Q feature. This predicts that embedded questions with wh-in situ should be impossible, in accord with the data reported in that work. However, as we showed in the introduction, there is an intervening factor in Boskovic's data which involves the choice of embedded complementizer. Once the appropriate complementizer is chosen, the embedded wh-in situ becomes good. Third, Boskovic's account entails that all examples of long-distance wh-in situ, involving those with non-intensional matrix verbs, are ungrammatical, since for Boskovic, all those examples would involve LF movement of the wh-phrase in situ. This movement, according to Boskovic, is so local that it cannot take place across a clause boundary. This analysis incorrectly rules out good instances of long-distance questions involving non-intensional verbs. Our analysis, in contrast, predicts the contrast in grammaticality between long-distance contexts involving intensional vs. non-intensional verbs. Taken in a more general perspective, our proposal that wh-phrases in French may be headed by an empty (silent) D head responsible for presuppositionality is not unprecedented. Rullmann and Beck (1998) make a similar proposal with regard to a different class of w/z-phrases, namely, which-phrases in English. They propose that these phrases have the structure roughly as in (32), which we compare with (24) repeated below: (32) [DP which [D' D NP]]
-English which-phrases
(24) [DP D [wh-phrase]]
-French wh-in situ
Our claim concerning structural similarities between French wh-in situ and English which -phrases is further strengthened given another similar aspect of their distribution. In particular, in both cases it is observed that the sentences in which such phrases occur are characterized as involving an existence presupposition originating in the wh-phrase. This state of affairs is expected given similar morphological makeups of both kinds of w/z-phrases; otherwise, it is a coincidence unaccounted for. 4.
Conclusion In this paper we movement and wh-in discussion. We have movement and wh-in
have investigated French questions involving whsitu. We have introduced a new range of data into shown that despite superficial appearance, the whsitu variants of French questions do not involve
70
CEDRIC BOECKX, PENKA STATEVA & ARTHUR STEPANOV
syntactic optionality, in particular, optionality of movement. Furthermore, we have demonstrated that no unified syntactic account of both types of whquestions is possible because there exist fundamental interpretive differences between them. These differences lie in the role of presupposition which is associated with the wh-in situ, but not with the fronted wh-phrase. We proposed that the difference in presuppositionality is reflected in the morphological make-up of wh-phrases in both cases. Specifically, we suggested that the wh-in situ is headed by a silent definite head D° while the w/z-phrase that undergoes movement is not. We explored the consequences of our proposal and showed that it adequately accounts for the entire paradigm of French wh-questions.
REFERENCES Beck, Sigrid & Hotze Rullmann. 1999. "A flexible approach to exhaustivity in questions". Natural Language Semantics 7.249-298. Boeckx, Cedric. 2000. "Properties of French interrogatives: A reply to Cheng and Rooryck". Ms., University of Connecticut. Boskovic, Zeljko. 1998. "LF movement and the minimalist program". Proceedings of NELS 28 ed. by P. Tamanji & K. Kusumoto, 43-51. University of Massachusetts, Amherst: GLSA. . 2000. "Sometimes in situ, sometimes in SpecCP". Step by Step: Minimalist essays in honor of Howard Lasnik ed. by Roger Martin, David Michaels & Juan Uriagereka. Cambridge, Mass.: MIT Press. & Howard Lasnik. 1999. "How strict is the cycle?" Linguistic Inquiry 30.689-701. Cardinaletti, Anna & Michael Starke. 1999. "Strong and deficient pronouns". Clitics in the languages of Europe ed. by Henk van Riemsdijk. Berlin: de Gruyter. Chang, Lisa. 1997. Wh-in situ phenomena in French. MA thesis, University of British Columbia. Cheng, Lisa. 1991. On the typology of wh-questions. Ph.D. Dissertation, MIT. & Johan Rooryck. Forthcoming. "Licensing wh-in-situ". Syntax. Chomsky, Noam. 1973. Conditions on transformations. S. Anderson & P. Kiparsky, eds. A festschrift for Morris Halle pp. 232-286. New York: Holt, Rinehart, & Winston. . 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. Hagstrom, Paul. 1998. Decomposing questions. Ph.D. Dissertation, MIT.
OPTIONALITY, PRESUPPOSIION, AND WH-IN SITU IN FRENCH
71
Heim, Irene. 1982. The semantics of indefinite noun phrases. Ph.D. Dissertation, University of Massachusetts, Amherst. --. 1992. "Presupposition projection and the semantics of attitude verbs". Journal of Semantics 9.183-221. Karttunen, Lauri. 1974. "Presupposition and linguistic context". Theoretical Linguistics 1.181 -194. Lahiri, Utpal. 1991. "The semantics of questions and the quantificational variability effect". MITWPL 13.163-178. Roberts, Ian. 1992. Verbs and diachronic syntax. Dordrecht: Kluwer Academic Press. Rullmann, Hotze & Sigrid Beck. 1998. Pressuposition projection, and the interpretation of which-questions. In Proceedings of SALT VIII ed. by D. Strolovitch & A. Lawson, 215-232. Cornell University. Tsai, Wei-Ten Dylan. 1994. On economizing the theory of A-bar dependencies. Ph.D. Dissertation, MIT.
WORD ORDER SHIFT AND NATURAL L2 ACQUISITION IN A PORTUGUESE CREOLE* J. CLANCY CLEMENTS Indiana University 0. Introduction In certain language contact situations, the degree of contact may become so intense that the typology of the non-dominant language can be entirely changed. In this study, I will discuss such a contact situation, focusing on how this typological shift has taken place. Moreover, I will also look at how speakers of this changing language have acquired the word order in their naturally learned second language, and more crucially, how to account for these processes. The question of typological shift in a given language has come to be understood as competition between grammatical subsystems (cf. Travis 1983, Dunn 1988, Kroch 1989, Santorini 1993, Taylor 1994). Expressed in general terms, when a surface change in a structure takes place incrementally, it may reflect a change in the underlying grammar that licenses the structure (Kroch 1989:239). The creation of a new licensing mechanism of a structure would then compete with the licensing mechanism already in place. Santorini (1993) and Taylor (1994) study the competition between verb-final and verb-medial position in the transformation of Yiddish and Ancient Greek from OV to VO languages. I adopt the same general model to account for the verb-medial (i.e. head-initial) to verb-final (i.e. head-final) shift evident in the data of the Portuguese-based creole being studied. Specifically, I will attempt to gauge the extent to which the competition model is able to capture: 1) the swiftness of the * I am grateful to the National Endowment for the Humanities (through the American Institute of Indian Studies, Supporting Grant FR 00615700) and to Fundaçao Calouste Gulbenkian, Portugal, for supporting this research. Thanks go also to Julie Auger for her insights on the study, and to the members of the LSRL audience for their comments. All errors and infelicities are my own.
74
J. CLANCY CLEMENTS
typological shift, especially its intermediate phases, and 2) the manner in which the speakers of this creole approximate the target language word order of their second language. It turns out that there is good evidence, not only for assuming independent, head-initial to head-final shifts for VO => OV, VAdjunct => AdjunctV, and Preposition => Postposition respectively, but also for arguing that, in the case of rapid, contact-induced change, the competition takes place simultaneously, though at different rates, at the level of constituents (a relatively low level), as well as at the level of the overall grammar. In this study, I will focus on the low-level competition, assuming that it takes place within one and the same grammar. The evidence indicates that, apart from positing the parameters "head-initial" arid "head-final", one would need to include "headinitial & head-final" as well. By adding a third possibility, the approach to be developed is able to model the two extreme stages, one head-initial and the other head-final, as well as the phase in which there is little to no discrimination between the two orders, i.e. "head-initial & head-final". 1. Recent studies on typological shift In the last 15 years, several studies have appeared on the topic of shift of object order relative to the verb, namely Dunn (1988) and Taylor (1994) for Ancient Greek, Travis (1983) for Chinese, Kroch for Old English (1989), and Santorini (1993) for Yiddish. Such research, however, does not address directly what occurs in the ordering of certain adjuncts relative to the verb or adpositions relative to their NP in a head-initial to head-final shift. In the studies most pertinent to the present one, Santorini shows that, at one point in its history, Yiddish had simultaneously an INFL-medial as well as an INFL-final structure, and Taylor demonstrates that Ancient Greek, around 450 B.C.E., possessed both a verb-medial and a verb-final structure in the same stage as well. These studies are based on written documents of the languages in question, and deal with cases of relatively gradual, not overtly contact-induced head-final to head-initial shift over a period of hundreds of years. The basic assumption of both Santorini and Taylor follows that of Kroch (1989), which is that syntactic change may involve competition between grammars. Taylor (1994:2) elaborates: "incremental linguistic change is accomplished by the gradual replacement of one grammatical system (or subsystem) by another. While the change is in progress, speakers can make use of forms from either grammar, but as time passes, a higher and higher percentage of the output is derived from the innovative grammar at the expense of the one it is replacing."
L2 ACQUISITION IN KORLAI
75
2. Korlai Creole Portuguese (KL) and its speakers The data used for this study were collected from KL, an Indo-Portuguese creole language spoken by approximately 750 speakers in Korlai, a village located around 200 kms. south of Mumbai on India's western coast. The Portuguese first arrived in the Korlai area in 1505, and already by 1530 there is evidence of considerable miscegenation in the population (Clements 1996:7-12). KL appears to have formed within a period of about 20 to 25 years after the Portuguese first landed in the area. Initial speakers of the creole were low-caste native Indians, slaves of the Portuguese, who subsequently converted to Christianity. By becoming Christians, they took on not only a new religion, but a new language and at least a portion of the Portuguese culture, as well. As members of the lower castes, they were already isolated from higher caste people. After conversion, however, because of their new religion and all it entailed, they also became isolated from their erstwhile low-caste Hindu peers. These barriers of caste and religion have had a strong isolating affect on KL up until the present century (cf. Clements 1990,1996 for details; on caste barriers, see Boxer 1963, 1975). Since the inception of the village around 1600 (Andrade 1945:18), the Korlai inhabitants have been farmers. Initially, we can assume (Teodonio D'Souza, p.c.) that they were tenant farmers, working for the Portuguese, and after the Maratha occupation of the area in 1740, for the Marathas. After 1947, the Tenants' Act was implemented, thanks to which the villagers were able to own their own land. Today, the vast majority of the Korlai inhabitants still are farmers by profession and many still own their own land. From 1740 onwards, Korlai villagers have had minimal contact with the Portuguese and their language, with the exception of the Portuguese-speaking priests. From 1964, all religious services and instruction have been held in Marathi, and the priests are native Marathi speakers. Regarding education, a four-year primary school was begun in 1916, and then in the 1980s, it was expanded to include ten grades. Census statistics from 1980 reveal that 51.7% of the adults over 18 had at that time one to four years of school, 36.6% more than four years of school, and 11.7% had never attended school. By contrast, in 1980, 86.7% of the children 3-18 were in kindergarten or attending school. Although KL has been in contact with Marathi (SOV), the regional language, since it formed, it is only in the last century that KL has begun to show signs of the strong cultural and linguistic pressure from Marathi. This influence is clearly
76
J. CLANCY CLEMENTS
visible in the structure of KL, and it has been documented in the speech of KL speakers. 3. The study The purpose of this study is to attempt to measure this increasing pressure of Marathi on KL. For this particular study, I examine head-initial to head-final word order change, focusing mainly on two aspects of the change: (1) The shift in the order of object NPs (i.e. DOs and IOs) relative to the verb, where by NP I am referring to full object NPs as well as object pronouns1, and (2) the shift in the order of locative, goal, instrumental, and sentential adjuncts relative to the verb. Temporal or manner adverbials were disregarded, as these have a much larger set of placement possibilities in KL. 3.1 Data collection I collected narratives —in all cases but two these consisted of fairy tales— from 16 native KL speakers and five native Marathi speakers. The stories were collected in the village, either in the speakers' homes, or at the researcher's residence in the village. The subjects were asked to tell any stories they knew and these were recorded by the researcher. In addition to telling stories in KL, the Korlai children also told stories in Marathi. In the case of the story retelling task, five KL children were told a story in Marathi and asked to retell the story in KL and then in Marathi. The purpose of this task was to see, given a controlled input, what their outputs would be in KL and Marathi. In the transcribed stories, I counted only declarative main clauses and direct quotes. Because of their different word order characteristics, no clauses containing interrogatives, imperatives, or dative subject constructions were taken into account. For the same reason, copulative constructions with predicate nouns/adjectives were also disregarded. The corpus proper consists of 1443 tokens from KL and 887 tokens from Marathi.
In KL, object pronouns carry stress and are, with the exception of 3s, bisyllabic. With respect to word order, these behave like full NPs. (i) Singular Plural 1 pari "me" pOnO "us" 2 pOrO "you" pudzo "you all" 3 pel "him, her" pelo "them"
L2 ACQUISITION IN KORLAI
77
3.2 Grouping of speakers Upon analyzing the data, one group of stories, told by one individual (an older woman), set themselves apart from all others because they contain features that do not form part of what constitutes KL today. It turns out that these are stories which the woman, as a young girl, learned from her mother-in-law and memorized. When she tells the stories, she reproduces them virtually the same each time. This was discovered upon requesting a second telling of one story a year after the first telling. These data constitute the first stage of the shift. With this small clarification, I present a characterization of the different speakers whose speech constitutes the four stages of the head-initial to head-final shift. In Stage 1, we have the data from the aforementioned woman. Two of her stories were obtained by other researchers (Theban 1985, Jackson 1987). The two versions of the same story told on different occasions in 1986, were also used. As stated, the stories of this speaker are memorized fairy tales. After her marriage in 1951, at the age of 19, she left Korlai for Mumbai, and has been in Mumbai ever since, although she maintains regular contact with relatives in Korlai. These stories reflect KL as it was used in stories by some older villagers around that time. To put a date on it, it would be KL of around 1900-1910.2 Stage 2 consists of a long narrative of a speaker who at the time of the recording (1986) had been out of the village for 46 years. His narrative tells of how famine hit Korlai in 1940, when he was 10 years old, how he made his way to Mumbai, found a job and established himself in the city. His would be KL from about 1930-1940. Stage 3 is comprised of speech from a cross section of three older KL speakers, who were around 65, 73, and 63 at the time of the recordings in 1986-88. Except for one narrative, these stories are all fairy tales. Stage 4 is made up of the speech of eleven children from seven to 16 years of age, taped in 1987-88, 1990-91, and 1995. All these narratives are fairy tales, told in KL and Marathi. Finally, for the purpose of comparison, I have included a control group, consisting of speech, also fairy tales, from five native Marathi speakers: four children, recorded in 1995, and one adult, recorded in 1991. All KL speakers in this study learned KL as their first and Marathi as their This date is calculated by the age of the mother-in-law, who in 1951 was around 50 years old and thus born about 1899. Her language, then, is a reflection of how people spoke during the first ten years of her life (cf. Labov 1981:180-81). Thus, the speech of the stories constitutes an earlier KL variety than that spoken by the person representing Stage 2. I consider this difference to be roughly 40 years.
J. CLANCY CLEMENTS
78
second language, starting L2 acquisition at four years old for the speakers 12 and under, and at six or seven years old for the speakers over 12. 4. Results Table 1 shows the token numbers and percentages of Object and Adjunct order relative to the verb in the four stages. This is displayed in graph form in Figure 1. At the first documented stage of the shift (i.e. the memorized stories), we find 80% VO and 90% VAdjunct order. At this point, KL is clearly a head-initial language, though the typological change has begun to occur. By the second stage, represented by the long narrative of the villager living outside of Korlai since 1940, KL is at the midway point for OV order. That is, it exhibits twice as much OV order (from 20% to 49%) at this point, with no significant change in the Adjunct order relative to the verb (from 10% to 8%). At stage three, however, we find that older KL speakers who have lived all their lives in Korlai favor OV order over half the time (58%), and AdjunctV order has risen dramatically, from 8% to 44%o. The fourth and final stage is exemplified by the data from 11 KL-speaking children, ages 7-16. These reflect a clear preference for OV order (73%), whereas AdjunctV order, at 47%), has not changed appreciably. This discrepancy between Object and Adjunct order relative to the verb does not seem to be attributable to Marathi, the contact language. ORDERS
VO
OV
VAdjunct
AdjunctV
Stage 1
8 0 % (101/127)
2 0 % (26/127)
90% (44/49)
10% (5/49)
Stage 2
5 1 % (83/163)
4 9 % (80/163)
9 2 % (99/108)
8% (9/108)
Stage 3
4 2 % (127/300)
5 8 % (173/300)
5 6 % (75/133)
4 4 % (58/133)
Stage 4
2 7 % (209/767)
7 3 % (558/767)
5 3 % (361/676)
4 7 % (315/676)
Table 1. The four stages of VO-> OV and VAdjunct-> AdjunctV shift. VO => OV: df 3, chi-square = 144, p < 0.001. VAdjunct > AdjunctV: df3, chi-square = 72.2, p < 0.001. The chi-square critical value for both is 16.226.3
3 Given that our chi-square values (144 for VO-OV, and 72.2 for VAdjunct-AdjunctV) are much higher than the critical chi-square value (16.266) the data are highly significant for the populations they represent.
L2 ACQUISITION IN KORLAI
79
Figure 1. The four stages of VO -> OV and VAdjunct -> AdjunctV shift in KL. (VO => OV: df3, chi-square = 144, p< 0.001. VAdjunct => AdjunctV: df3, chi-square = 76, p ≤ 0.001) In a control study of five native L1 Marathi speakers (cf. the upper right hand corner of the graph in Figure 1), in the data which consist of 416 declarative noncopulative, finite-, main-clause tokens, OV order is at 95% and AdjunctV order at 92%. Thus, it seems safe to conclude from Figure 1 that the head-final shift for objects and adjuncts takes place in parallel, but independently. To test the extent to which KL object vs. adjunct order in the children's speech affected word order frequencies in their second language, Marathi, word order in the children's KL and Marathi stories was contrasted. Table 2 shows the tokens and percentages, and Figure 2 includes Marathi monolingual word order, in the upper right-hand corner, as well as the orders in KL and Marathi of the Korlai children. ORDERS
VO
OV
VAdjunct
AdjunctV
Korlai Children (KL)
27% (209/767)
73% (558/767)
53% (361/676)
47% (315/676)
Korlai Children (Marathi)
15% (17/114)
85% (97/114)
23% (94/415)
77% (321/415)
Table 2. VO => OV and VAdjunct => AdjunctV for KL (df 1, chi-square = 102.84, p< 0.001); VO => OV and VAdjunct => AdjunctV for Marathi (df 1, chi-square = 3.23, p< 0.10) in Korlai children's speech (KL [L1] and Marathi [L2])
80
J. CLANCY CLEMENTS
As mentioned above, in their native KL, the children display a clear preference for OV order (73%), whereas AdjunctV order, at 47%, shows virtually no preference for head-initial or head-final position. Revealingly, when speaking in Marathi, which is heavily OV, Korlai children exhibit a higher percentage of OV and AdjunctV order.
Figure 2. VO => OV and VAdjunct => AdjunctV shift in children's speech (KL and Marathi) In fact, their Marathi OV order, at 85%, is very near to 84% average of the percentages of their KL OV order at 73%), and that of the Marathi control group at 95%). Similarly, they have 11% AdjunctV order in their Marathi, which again is slightly higher than the average of 70% between 47%) KL and 92%o Marathi Adjunct order relative to the verb. Based on these data, it is apparent that there is not complete transfer of KL word order frequencies to Marathi, nor do the Korlai children exhibit native-speaker word order frequency in Marathi. Rather, with respect to Object and Adjunct order relative to the verb, the KL children's Marathi has an interlanguage quality in that it is an approximation to Marathi native word order frequency. Important to note in this interlanguage word order is that OV and AdjunctV orders are kept separate, as they are in the children's KL. Yet, in both cases, there is an approximation toward the native Marathi word order. This confirms that there are two levels of competition present, reflecting processes that are developing at a different pace, albeit in tandem. Part of the data displayed in Table 2 and Figure 2 is taken from a storyretelling task, which by itself yielded interesting results. Several KL-speaking
L2 ACQUISITION IN KORLAI
81
children were told a story in Marathi by a native Marathi speaker, and were asked to retell the story, first in KL and then in Marathi. Table 3 gives the input orders, as well as the orders in KL and Marathi. These statistics are plotted in Figure 3. ORDERS
VO
ov
VAdjunct
AdjunctV
Marathi Story (input)
3 % (2/39)
97% (37/39)
12% (5/40)
88% (35/40)
Story Retell in KL
19% (10/53)
81% (43/53)
73% (44/60)
27% (16/60)
Story Retell in Marathi
23% (13/57)
77% (44/57)
53% (31/59)
47% (28/59)
Table 3. Story retelling: Input, KL, and Marathi
Figure 3. Story retelling by KL-speaking children: VU => UV and VAdjunct => AdjunctV orders of input, KL, and Marathi
Although the raw numbers are not as robust as those in Table 1, we still see that KL-speaking children, when retelling in KL a story they have heard in Marathi, make a clear distinction between Object and Adjunct order relative to the verb. In fact, they favor head-final order in KL even more strongly in the retelling exercise (73% in Figure 2 vs. 81% in Figure 3), while at the same time preferring head-initial VAdjunct order much more than in Marathi (47 % AdjunctV order in Figure 2 vs. 27% AdjunctV in Figure 3). In their Marathi
J. CLANCY CLEMENTS
82
retelling of the story, they again show the stronger preference for OV order (77%) vs. AdjunctV order (47%), not unexpected for the KL-speaking children based on the data from their own language. Again, this is evidence that there are two shifts going on simultaneously, but at different rates. 5.
Discussion Based on the data presented in the previous section, one can posit a number of crucial phases in the evolution of a rapid head-initial-to-head-final shift. There are two developments, displayed in Table 4, where a beginning headinitial phase gives way to one in which there is a major and a minor word order. The next phase is then indeterminate, followed by minor-major phase, which finally gives way to a sole head-final phase. Although, both the VO => OV and VAdjunct => AdjunctV shifts currently going on in KL correspond to phases 2, 3, and 4 in Table 4, they are at different phases in the shift process. The data suggest that the AdjunctV shift seems about one phase behind the OV shift, although this is not exactly accurate, since we saw above in Figures 2 and 3 that AdjunctV order in KL children is at 47% and 27% respectively, i.e. not yet decisively at the "head-initial & head-final" phase. 1
2
4
VO
VO major ov minor
VO OV
vo minor OV major
VAdjunct
VA major av minor
VA AV
3
5
6
(OV) (va minor AV major)
(AV)
Table 4: Phases of VO => OV, and VAdjunct => AdjunctV shift Given that the notion of parameter allows for binary options only, and not "major" or "minor" word order options, it is difficult to model all these phases of the shift within a theory of competition between grammars. Recall that this approach routinely uses two, although there are four logical possibilities: headinitial, head-final, head-initial & head-final, and neither head-initial nor headfinal. We will see that the data strongly support at least the first three possibilities. In what follows, I will assume, and not motivate, the head-initial and head-final parameters. However, I will address two issues that pertain to the third possibility and ways to motivate it within a parametric model: First, the difference between Adjunct and Object order relative to the verb, and second, certain KL doubling constructions and what they might add to the question of the third possibility.
L2 ACQUISITION IN KORLAI
83
As seen above, relative to the verb, Adjunct and Object order shift operate in tandem, although on separate axes, as it were. A minimally simple way to represent this in terms of phrase structure would be to posit the Adjunct as belonging to a different constituent than the object NP. Given the data in Figures 2 and 3, it is reasonable to argue that, in current KL, V' - Adjunct is head-initial and NP - V is head-final, as in (l). 4
Assuming some structure like (1), then, it would not seem accurate to maintain that the head-initial V-Adjunct order is in direct competition with the head-final NP-V order, since the data suggest that they are developing separately from one another. However, it is reasonable to adopt the idea that there are two local competitions in play: one involving Adjunct order, the other involving Object order relative to the verb. If this is assumed, it follows that in the grammar of KL-speaking children, both head-initial and head-final parameters can exist independently, relative to the verbal head. This is the first argument for positing in UG three parametric possibilities for order, i.e. head-initial and headfinal, along with head-initial & head-final.5 Another set of data that favor positing the third possibility is the phenomenon of doubling, of the type VOV and PNPP. In KL one finds verb doubling commonly in the speech of both older and younger speakers alike. Examples are shown in (2). (2)
a.
El tada sikwet tada, mes s/he gives fifty gives month "S/he gives fifty [rupees] a month."
su. GEN (Rose, 65+ years)
4 In a phrase structure such as (2), there is no straightforward way of representing the indeterminate phase of the shift. This would arguably have to be stipulated. 5 It is well known that Chinese has VO but AdjunctV word order. Thus, it is not surprising that in KL Object and Adjunct order relative to the verb would change separately of one another.
84
J. CLANCY CLEMENTS
b.
El Ajay ki te, el tada pastis rhiipi peret tada. he Ajay who is he give fifty rupee back give "Him, Ajay, right, he gives back the fifty rupees." (Sharmila, 10 years)
This type of doubling is also common with certain adpositions in KL. Although KL has become largely a postpositional language, both younger and older KL speakers still use prepositions. So, for example, both younger and older speakers routinely produce utterances such as those in (3) and (4). (3)
(4)
a.
bash alb under tree b. alb su bash tree GEN under "under the tree" det kadz su det in house GEN in "in the house"
The examples in (5) are taken from a story told by a 60+ year old man. (5)
Gonta ichew ani picho rhib bhur bhur su rhib bags filled and threw on donkey donkey GEN-on "He filled the bags and threw them [distributively] on the donkeys."
Arguably, doublings such as the one in (5) reflect a transitional phase between one grammatical subsystem and another in a rapid shift. I would like to suggest that cases such as (5) are not merely instances reflecting that there are two subsystems in competition, but rather, since both can be applied at the same time to the same constituent, that they are part of the same grammar of the speaker. That is, in one and the same verb, there is bidirectional case assignment. This seems less likely to take place in an either-or-type competition between two grammatical systems. Rather, bidirectional case assignment would become a feature of a single grammar.6 If one looks further, one finds such cases of doublings in other contact situations as well. For example, Minaya and Luján (1982) and Luján, Minaya Escobar (1992) finds adposition doubling in the native Quechua speakers learning Spanish as a second language. In Japanese-English codeswitching, Yasuhiro Shirai (p.c.) has encountered adposition doubling as well.
85
L2 ACQUISITION IN KORLAI
and Sankoff (1984) found that when children whose first language is Quechua learn Spanish, they impose the syntax of their native language on the second language. Only the verb-object word order is given here. Note in Table 5 below that, at age five, when the children begin to learn Spanish, we find not only VO and OV orders in virtually equal percentages, but also VOV doubling 25% of the time. This evidence of bidirectional case marking decreases at age seven by 9%, and by age nine, by again 9%. However, the fact that it occurs so robustly implies such case marking must be a part of the grammar of these children.
Number of Subjects
Age
OV
VO
VOV
3
5
38% (192)
37% (186)
25% (124)
100% (502)
3
7
32% (168)
52% (273)
16% (85)
100% (526)
3
9
26% (141)
67% (365)
7% (41)
100% (547)
Total
Table 5. Quechua-speaking child learners of Spanish: OV-> VO, df 4, chi-square = 105.75, p ≤ 0.001. Chi-square critical value is 18.467. Thus, it seems that in certain language-contact situations, in one and the same grammar of a speaker, we find verbs that can possess simultaneously head-initial as well as head-final properties, i.e. bidirectional case assignment. One economical way to account for it is to allow the "head-initial & head-final" option, along with the other much more common ones. It might be that this third possibility is only accessed by speakers who are experiencing a rapid typological shift in their language, or are learning a language that is typologically diverse from their own. In both situations, there is good evidence for the presence of the "head-initial & head-final" possibility. As a last observation, I would like to add that this doubling phenomenon is also found in morphology. In Puerto Rican Spanish, for example, the verbal suffixal marker for the 3pl, -n, is optionally doubled in the imperative, as shown in (6a). The standard Spanish form is given in (6b). (6)
a. b.
Siéntensen sit-IMPER-3pl-REFL-3pl "Sit down!" Siéntense sit-IMPER-3pl-REFL "Sit down!"
86
J. CLANCY CLEMENTS
Arguably, the doubling in (6a) marks a transitional phase in the process at the end of which the reflexive clitic pronoun se will appear inside the 3pl marker -n. At the end of the change, we will be able to say that the status of se will have changed from that of a clitic to that of a bona fide suffix. The question about competition of grammars in this case then becomes: does the person who utters the form in (6a) have competing grammars, or a grammar in which se has an indeterminant status? I suggest the second option because it is simpler than positing two separate grammars. And by Occam's razor, the simpler solution to a problem wins out over the more complex one. Following the same line of argument for Object/Adjunct order relative to the verb in KL, it seems that it is simpler to assume that the question is one of case assignment rather than the competition between two separate grammars. Santorini (1993) and Taylor (1994) do not speak of competition between grammars, but rather of competition between grammatical subsystems. I would like to go one step further and suggest that directional case assignment may simply become bidirectional within one and the same grammar of a set of KL speakers. In such a phase, the grammar of these speakers would have both headinitial and head-final parameters. Given the evidence from Adjunct vs. Object order relative to the verb seen above, as well as the evidence regarding doubling of objects and adpositions, there appears to be a need to posit, along with the head-initial and head-final, also head-initial & head-final as well. Although the recognition of this third option may add complexity to the theory, I consider it a just price to pay for being able to model more accurately a situation which is most likely not that uncommon in the languages of the world.
REFERENCES Andrade, Franciso de. 1945 [1594]. Comentários da victória de Chaul ed. by Jorge de Faro (ed.). Colecção Pelo Império 109. Lisbon: Divisão de Publicações e Biblioteca, Agéncia Geral das Colónias. Boxer, Charles. 1963. Race Relations in the Portuguese Colonial Empire. Oxford: Clarendon. . 1975. Women in Iberian Expansion Overseas, 1415-1815. Oxford: Oxford University Press. Clements, J. Clancy. 1990. "Deletion as an indicator of SVO => SOV shift". Language Variation and Change 2.103-33.
L2 ACQUISITION IN KORLAI
87
. 1996. The genesis o f a language. Amsterdam & Philadelphia: John Benjamins. Dunn, Graham. 1988. "Syntactic word order in Herodotean Greek". Glotta 66.63-79. Escobar, Anna María. 1992. "El español andino y el español bilingüe: semejanzas y diferencias en el uso del posesivo". Lexis 16.189-222. Jackson, . David. 1987. "Um conto folclórico no crioulo Indo-Português". Bulletin des Etudes Luso-Brésliennes 89-98. Kroch, Anthony S. 1989. "Reflexes of grammar in patterns of language change". Language Variation and Change 1.199-244. Labov, William. 1981. "What can be learned about change in progress from synchronic description?" Variation Omnibus ed. by David Sankoff & Henrietta Cedergren, 177-199. Edmonton: Linguistic Research, Inc. Luján, Marta, Liliana Minaya, and David Sankoff 1984. "The universal consistency hypothesis and the prediction of word order acquisition stages in the speech of bilingual children". Language 60.343-71. Minaya, Liliana and Martha Luján. 1982. "Un patrón sintáctico híbrido en el habla de los niños bilingües en quechua y español". Lexis 6.271-293. Santorini, Beatrice. 1993. "The rate of phrase structure change in the history of Yiddish". Language Variation and Change 5.257-84. Taylor, Ann. 1994. "The change from SOV to SVO in Ancient Greek". Language Variation and Change 6.1-37. Theban, Laurentiu. 1985. Situaçao e perspectivas do Português e dos crioulos de origem portuguesa na India e no Sri-Lanka." Congresso sobre a situação actual da lingua portuguesa no mundo. Actas, vol. 1, 269-86. Lisbon: Instituto de Cultura e Lingua Portuguesa. Travis, Lisa. 1983. "Word order change and parameters". MIT Working Papers in Linguistics 5.277-89.
UNACCUSATIVITY AND THE IMPERSONAL CONSTRUCTION OF FRENCH SARAH CUMMINS Université Laval
0.
Introduction This paper examines unaccusative mismatches in the Impersonal Construction (I ) of French. I show that no explanation for these mismatches that maintains the Unaccusative Hypothesis (UH) is tenable. The evidence instead leads to the conclusion that single-argument verbs map freely to unaccusative structures; thus the UH is disproven and the broader principle of lexically-driven mapping is challenged. I argue that a system of free linking and postsyntactic compositional interpretation accounts for patterns of verb selection in the I . The UH (Perlmutter 1978) posits two classes of intransitive verbs: unergatives have an external (subject) argument, and unaccusatives have an internal (object) argument. According to Levin & Rappaport Hovav (1995:4), unaccusativity is "semantically determined" (by the semantic relationship between verb and argument) and "syntactically represented" (by mapping the argument to object position). The I considered here is exemplified in (1). (1)
a. Il est arrivé trois princesses. it is arrived three princesses "There arrived three princesses." b. Il dort un chat au coin de la cheminée. (Hulk1989:59) it sleeps a cat at-the corner of the chimney "There is a cat sleeping at the corner of the chimney."
These examples illustrate the following characteristics of the I : its grammatical subject is the expletive pronoun il (masculine singular nominative); the verb agrees with il in number and, in the case of participial agreement, gender; the construction has a locative reading, even in the
SARAH CUMMINS
90
absence of an overt locative; and the Defmiteness Effect (DE) is observed on the postverbal NP (PVNP), typically marked as indefinite. The 1 , then, appears to be unaccusative in structure, and it is indeed taken as a diagnostic of unaccusativity. The UH is faced with the question of how apparently unergative verbs, such as dormir "sleep" in (lb), appear in an apparently unaccusative structure—why the argument of an unergative verb is mapped to object position. Three types of explanation have been proposed for such "unaccusative mismatches"; in the following section I examine each in turn. 1. Explanations for unaccusative mismatches 1.1 Unergative verbs in the I constitute an "exception " The presence of unergative verbs in the I has been qualified as "marginal" or "exceptional" by some researchers (e.g. Labelle 1992; Herschensohn 1982, 1996; Hulk 1989). Corpus evidence offers a means of evaluating this claim. In his work on various impersonal constructions in French, Hériau (1980) collected 14,802 tokens of the I at issue here. He found that three verbs (être "be", rester "remain", and exister "exist") account for 54% of tokens; the next three most frequent verbs (venir "come", arriver "arrive, happen", manquer "be lacking") account for a further 18% of tokens. These figures are consistent with the claim that unaccusative verbs are the most "natural" in the I . However, Hériau's list of the 50 most frequent verbs in the I (which account for 96.13% of tokens) includes several apparently unergative verbs, e.g. courir "run", rôder "prowl", sonner "ring". Moreover, many other unergative verbs are attested in the I , albeit not frequently; examples, from diverse semantic classes, are shown in (2). (2)
briller "shine" crier "cry out" galoper "gallop" pleurer "weep" retentirV'resound"
cahoter "bump along" danser "dance" luire "gleam" respirer "breathe" saigner "bleed"
chanter "sing" diner "dine" nager "swim" sauter "jump" siffler "whistle"
The corpus evidence shows that unergative verbs are not the most frequent in the I , but they are nonetheless common. Some are in fact more frequent than certain unaccusative verbs. Their presence cannot be dismissed as marginal, exceptional, or unusual. Unaccusativity is neither a necessary nor a sufficient condition for a verb's appearance in the I , and a further explanation for these unaccusative mismatches is required.
UNACCUSATIVITY
91
1.2 The I is not unaccusative A second possible explanation is the claim that the structure of the I is not actually unaccusative (at least when it contains an unergative verb) and thus has no bearing on the UH (cf. Levin & Rappaport Hovav 1995; Safir 1985). The correct word order can be achieved by mapping under the UH, with a VP-internal subject and verb movement to T, as in (3b) and (4b). Alternatively, the argument of an unergative verb could be extraposed (4c). (3)
a.
Il arrivera un malheur. it will-happen an unhappiness "Something bad will happen."
(unaccusative)
(4)
a.
Il y dormait un it there was-sleeping a "A cat was sleeping there."
(unergative)
chat. cat
Although the correct word order can be obtained, the question of Case is problematic. In non-IC sentences, the argument of both unaccusative and unergative verbs appears in subject position and receives nominative Case in relation with T/INFL. The Case of the PVNP in the I , however, is not evident.
SARAH CUMMINS
92
Three solutions have been put forward. Belletti (1988) proposes that this argument receives inherent partitive Case from the unaccusative verb. While this proposal has the advantage of accounting for the DE, it assumes that unergative verbs do not appear in the I and offers no solution for such instances. Herschensohn (1996) proposes that all PVNPs in the I receive structural partitive Case, an object Case that can be assigned when accusative Case fails. This solution assumes that the arguments of unaccusative and unergative verbs are alike in the I ; thus it argues against the UH. A third solution, following Safir (1985), holds that the PVNP of an unergative verb receives nominative Case in association with the expletive subject. Under this analysis, the PVNP enters into a Case-chain with il; the two are co-indexed and both bear nominative Case. According to Safir, elements in such an unbalanced theta-chain manifest the DE. This proposal maintains the UH by allowing for different structures for unaccusative and unergative verbs. The relevant structure with an unergative verb would be as in (5). (5)
a.
Il paissait sur la colline deux ou trois brebis. it was-grazing on the hill two or three sheep "There grazed on the hill two or three sheep."
This solution, too, runs into problems. Specifically, it clashes with evidence that the PVNP is in object position and receives object Case, as Herschensohn (1996) proposes. The first type of evidence comes from the clitic en. The PVNP in the I may be the associate of this clitic in its partitive meaning. (6a) and (b) correspond to (la) and (b) respectively. (6)
a.
Il en est arrivé trois. it PRT is arrived three. "There arrived three."
UNACCUSATIVITY
b.
93
Il en dort un au coin de la cheminée. at-the corner of the chimney it PRT sleeps one "There is one sleeping at the corner of the chimney."
These examples show that the postverbal associate of en need not be marked with an overt preposition. The only position for such an unmarked associate is the direct object position. In particular, extraposed arguments may not be associates of en, whether they are underlying external arguments, as in (7) or underlying internal arguments, as in (8) (from Pollock 1986:221). (7)
a.
b.
(8)
a.
b.
Devant la maison stationnaient plusieurs voitures. before the house parked several cars "In front of the house were parked several cars." *Devant la maison en stationnaient plusieurs. before the house PRT parked several "In front of the house were parked several." Quand a été imprimé un faux billet? when has been printed a false note "When was a counterfeit bill printed?" *Quand en a été imprimé un? when PRT has been printed one "When was one printed?"
If arguments of unaccusatives in the I are in direct object position but arguments of unergatives are not, it is expected that only arguments of unaccusatives would be potential associates of en. In fact, en relates to the postverbal argument of both verb classes in the I , as shown in (6). Evidence from the I with a transitive verb, as in (9), provides a further indication that the PVNP is in direct-object position. (9)
a.
Il vous guette des renards. (627)1 it you watches of-the foxes "There are foxes watching you." b. Il t' attend un pays ruisselant de lait et de miel. it you awaits a land flowing of milk and of honey "There awaits you a land of milk and honey." (626)
Unless indicated otherwise, page numbers following examples refer to Hériau (1980).
SARAH CUMMINS
94
If the PVNP in (9) is in an extraposed position, it should be possible for a lexical NP to fill the direct object position. But this is impossible: (10) a.
b.
*Il guette vos amis des renards. it watches your friends of-the foxes "There are foxes watching your friends." *Il attend ta mère un pays ruisselant de lait et de miel it awaits your mother a land flowing of milk and of honey "There awaits your mother a land of milk and honey."
The conclusion again is that the PVNP fills the direct object position. The I with a transitive verb also gives information about the Case of the PVNP. With canonical mapping of transitive verbs in non-IC sentences, the external argument receives nominative Case in relation to T/INFL and the internal argument receives structural accusative Case; a clitic object is accusative. In the I , if the PVNP bears nominative Case, we would expect accusative Case to be available and observable when the "internal" argument is realized as a clitic. This is not so. Accusative and dative clitics are distinguished only for third person: le/la/les vs. lui/leur. Distinctly accusative clitics are ungrammatical in the I . (11) a.
b.
*Il le/la/les guette des renards. it him/her/them-ACC watchesof-the foxes. "There are foxes watching him/her/them." *Il l'/les attend un pays ruisselant de it him-her/them-ACC awaits a country flowing of lait et de miel. milk and of honey "There awaits him/her/them a land of milk and honey."
The evidence shows that the direct object position is filled in the I , regardless of verb type, and object Case is assigned to this position, regardless of verb type. The arguments of unaccusative and unergative verbs are the same, filling the direct object position and receiving Case by virtue of this position. The verb in the I has an object argument and no subject argument. The claim that the I with unergative verbs is not an unaccusative structure and has no bearing on the UH cannot be maintained.
UNACCUSATIVITY
95
1.3 Variable behaviour accounts A third type of explanation for unaccusative mismatches is the proposal that certain well-defined semantic classes of unergative verbs undergo a semantic shift enabling them to map as unaccusatives (cf. Levin & Rappaport Hovav 1995). If this approach is applied to the I , the semantic change would be acquisition of the meaning of appearance or existence in the world of discourse (Guéron 1980:653-654). To illustrate, paître "graze" in (12a) would have the expanded meaning of "be present and grazing"; fourmiller "swarm" in (12b) would mean "exist in sufficient quantity to swarm". (12) a.
b.
Il y paissait deux ou trois brebis. it there was-grazing two or three sheep "There were two or three sheep grazing there." Ah! mais il fourmille des exemples. ah but it swarms of-the examples "Oh, but there are swarms of examples!"
(228)
One problem faced by this account is the difficulty of delimiting the classes of verbs displaying variable behaviour. No semantic class of unergative verbs (verbs of emission, movement, bodily functions, etc.) can be excluded. If all unergative verbs can in principle display variable behaviour, the distinction between unaccusatives and unergatives is nullified. Moreover, the prediction is that an unergative verb can undergo the lexical transformation and then appear in other unaccusative constructions. However, the meaning of existence/appearance is discernible only in the I . To illustrate, (13a) shows an unaccusative verb, rester "remain", whose argument has mapped to object position and then become subject. (13b) shows an unergative verb, courir "run", which has supposedly undergone the lexical transformation turning it into an unaccusative; its argument behaves in the same way. But the semantic shift is absent from (13b); there is no existence/appearance reading. (13) a.
b.
Deux two "Two Deux two "Two
enfants restaient dans la children remained in the children remained in the room." enfants couraient dans la children ran in the children ran around in the room."
salle. room salle. room
SARAH CUMMINS
96
Hoekstra & Mulder (1990) propose a different kind of variable behavior account. Under their analysis, unergatives can become "copular" verbs; the shift in meaning leads to a shift in mapping, with the unergative copular verb taking a Small-Clause complement, as in (14). (14) a.
Il crissait du cristal sous vos it crunched of-the crystal under your "Crystal crunched under your soles."
semelles. soles
Hoekstra & Mulder claim that unergatives appear in the I only if a locative PP is present. But this is not so; (15) shows examples of the I with a verb classified as unergative, yet without a locative PP. (15) a.
Il bourdonnait des conversations menées dans it buzzed of-the conversations led in toutes les langues du monde. (224) all the languages of-the world "Conversations buzzed, in all the languages of the world." b. Il chantait un artiste espagnol inconnu en France. it sang an artist Spanish unknown in France (Hoekstra & Mulder 1990:48) "A Spanish artist, unknown in France, was singing."
Yet the analysis requires a PP to anchor the Small Clause. If it is not present, there is no distinction in the structural requirements of unaccusative and unergative verbs. Since the two classes behave alike, and no restrictive, semantically coherent account can be given of variable behaviour, this explanation for unaccusative mismatches must also be rejected. In summary, this examination of the I has shown that its syntactic structure is unaccusative, with an internal but no external argument. Yet both unergative and unaccusative verbs map freely to this structure, without undergoing a lexical transformation. The UH is disproven. Moreover, the
UNACCUSATIVITY
97
principle of lexically-driven mapping, of which the UH is the cornerstone, is thrown into doubt. As an alternative to this principle, the null hypothesis is that of free-linking: that arguments map freely to syntactic positions. Section 2 presents an account of the verb-argument pairs found in the I based on free linking and postsyntactic compositional interpretation. 2. A free-linking account The Free Linking Hypothesis, formulated by Ghomeshi & Massam (1994:178) proposes that "any syntactic position can contain any semantic participant, subject to interpretability". Free linking is restricted by the Compatibility Constraint, requiring that "meaning contributed from a given source must be compatible with meaning contributed from all other sources". An examination of the meaning contributed by various sources and the compatibility among them sheds light on verb-argument pairs in the I . 2 2.1 Constructional meaning Constructional Grammar (cf. Goldberg 1995) holds that constructions themselves have meaning and that the meaning of a construction determines the relative frequency of verbs appearing in it. In Constructional Grammar terms, the meaning of the I would be expressed as BE < theme loc. The most frequent verbs in the I are indeed those whose meaning matches this constructional meaning: verbs of existence or presence—être "be", rester "remain", and exister "exist", which account respectively for 24%, 19%, and 11%) of the tokens in Hériau's corpus. Verbs whose semantics is close to that of the construction—i.e. manquer "be lacking" (negation of presence or existence) and venir "come" and arriver "arrive, happen" (coming into presence or existence)—are the next most frequent, accounting for 6% of tokens each. The hypothesis concerning verb frequency is confirmed. Construction Grammar treats the construction as an unanalysable whole. This view leaves implicit the reasons behind the association of a meaning with a construction and thus treats this association as arbitrary. In fact, it is possible to "deconstruct" the construction and examine the semantic contribution of each element, thereby providing some understanding of the basis of constructional meaning.
2
In this account, I assume that lexical arguments are those that are essential to the meaning of the verb. Lexical arguments are not obligatorily mapped to syntax, but interpretation is based first on lexical arguments.
SARAH CUMMINS
98
2.2 Meaning contributed by elements of the construction 2.2.1 Sentence shape and subject of predication. Saccon (1992) shows that the "subject of predication" must be outside the VP projection. In a presentational sentence like the I , the verb does not predicate of the PVNP; such sentences are interpreted as predicating of a location. The Compatibility Constraint limits the verb-argument pairs able to predicate of a location. 2.2.2 Expletive subject il and interpretation of the I . The fact that the I has an expletive subject also influences interpretation. First, the number and form of semantic arguments is limited in consequence (see Herschensohn 1996). Interpretation possibilities are also limited. I propose the following principle of postsyntactic interpretation: (16) Subject Initiation Principle (SIP): When a sentence depicts (i) an event and (ii) its initiation, an initiator argument must be in subject position. It is useful to consider the SIP the poststyntactic counterpart of presyntactic generalizations that have been proposed, such as Grimshaw's (1990:110) statement that "agents always map as subjects" and Levin & Rappaport Hovav's (1995) Immediate Cause Linking Rule, which places the "immediate cause" of an event in subject position. Like them, it identifies a correspondence between subject position and initiation. The presyntactic rules determine mapping and are based only on the semantic relationship between verb and argument. The postsyntactic SIP determines interpretation and takes other sentence elements into account. Before assessing the effect of the SIP on interpretation of the I , it is useful to see how it works with other types of sentence. The simplest examples are sentences depicting events and containing verbs with two lexical arguments. The SIP requires that the argument in subject position be interpreted as initiator, even with anomalous results (17a); if only one argument is mapped (to subject position), as in (17b), it is interpreted as initiator, even if the interpretation as patient would also be plausible. (17) a.
#La tomate a mangé Yvon. the tomato has eaten Yvon "The tomato ate Yvon."
UNACCUSATIVITY
b.
99
Manon a dérangé Manon has disturbed "Manon disturbed."
The same is true in eventive sentences containing pronominal verbs with two lexical arguments (cf. Zribi-Hertz 1982). (18) a.
b.
Je me transporte. I me transport "I transport myself." Les patrons se séquestrent. the bosses REFL sequester "The bosses are locking themselves away." #Un pantalon se lavera. a pants REFL will-wash "A pair of pants will wash itself."
However, if the sentence contains elements that contribute to a non-event interpretation, the SIP no longer holds. In (19), the adverbs facilement "easily" and toujours "always" and the generic pronoun ça contribute to a property reading and the subject of the sentence is not interpreted as initiator. (19) a.
b.
Je me transporte facilement. I me transport easily "I am easy to transport." Les patrons ça se séquestre? the bosses it REFL sequesters "Bosses can/should be locked up." Un pantalon se lavera toujours à a pants REFL will-wash always at l' eau enzymé. (Zribi-Hertz 1982:356) the water enzymed. "A pair of pants will always be washed in enzyme water."
3 I adopt Auger's (1994:271-293) analysis of ça as a neuter subject marker that blocks the sharing of certain features, such as agreement, between subject and verb.
SARAH CUMMINS
100
When the sentence depicts an event without depicting its initiation, the SIP is also suspended, as in passive sentences and in sentences with effected patient subjects of pronominal verbs in the "se-passif" use. Under the SIP, I sentences cannot depict initiation of an event. The reasons underlying the overall constructional meaning of the I are the requirement for a locative subject of predication and the inability to express initiation of an event. The I must therefore depict a "state of affairs": an ongoing state or activity or the aftermath of an event. The semantic classes of verbs compatible with this requirement become clearer. Verbs of existence do not represent an event and are eminently suited for depicting a state of affairs. Verbs of appearance/occurrence and weather verbs do depict events, but not their initiation; both classes are frequent in the I . Activity and movement verbs can be compatible to the extent that activity or movement coincides with presence or appearance. The more agentive a verb-argument pair is, the less likely it is to be compatible with the I . An agentive argument would demand to be interpreted as initiator (impossible under the SIP) and would be in competition with the locative to be the "subject of predication". 2.2.3 Tense. The Compatibility Constraint also offers an explanation for the verb tenses found in the I . Past perfective tenses, such as the passé composé, indicate that a state or an event came to an end in the past: (20) a.
b.
Céline a dormi dans ce Céline has slept in this "Céline (has) slept in this bed." Marie a été médecin. Marie has been doctor "Marie has been a doctor."
lit. bed
One would expect perfective tenses, then, to be incompatible with the I , which must depict a "state of affairs". Indeed, when a perfective tense replaces an imperfective in certain I sentences, grammaticality suffers: (21) a.
b..
*Il a dormi un chat au coin de la cheminée. it has slept a cat at-the corner of the chimney "There slept (perfective) a cat at the corner of the chimney." *Il fourmilla des exemples. it swarmed of-the examples "There were (perfective) swarms of examples."
UNACCUSATIVITY
101
However, some verbs do appear in perfective tenses in the I , as in (22).These are Transitions (Pustejovsky 1995)—complex events depicting a process culminating in a state. Tense perfectivity applies to the point of transition; the resultant state satisfies the need for a "state of affairs". Typical classes are verbs of appearance, occurrence, disappearance, and remaining. (22) a.
Il éclata, au début de l' été, it burst at-the beginning of the summer cette année- là, une épidémie meurtrière. (227) that year there a epidemic killer "There broke out, in early summer that year, a deadly epidemic." b. Il a persévéré dans la mort l' oubli it has persevered in the death the forgetting qu' il avait déjà connu pendant sa vie. (233) that he had already known during his life "There persevered in his death the oblivion he had known during his life."
2.2.4 Information status of the object. A final element in interpretation of the I is the information status of the object. Although the DE is widely observed, some PVNPs are definite; a few examples are given in (23). (23) a. Il vint aussitôt la pluie et la nuit. (802) it came immediately the rain and the night "There came immediately the rain and the night." b. Mais il restera la France. (805) but it will-remain the France "But there will remain France." il lui poussa la belle idée d'appeler le it him grew the fine idea of to-call the jeune homme qui causait avec Louise. (808) young man who was-chatting with Louise "There grew in him the fine idea of calling the young man who was chatting with Louise." Birner & Ward (1998), examining similar cases in English, conclude that the PVNP in such sentences must be discourse-new. Usually discourse-new NPs are indefinite, because they are not individuable by previous presence in the
102
SARAH CUMMINS
discourse. Felicitous definites involve discourse-new NPs that are individuable by some other means; they include discourse-new tokens of familiar types, as in (23a), or discourse-new entities with individuating descriptions (23c). extensive study of corpus data would be required in order to determine if the same principles hold for PVNPs in the I in French, but the approach seems more promising than Case-based solutions with obligatory morphological effects, which offer no explanations for definite markers on the PVNP. 3.
Conclusion None of the explanations put forward for unaccusative mismatches can be maintained in the case of unergative verbs in the Impersonal Construction. Unergative verbs are not unusual in the I , and their use does not constitute an exception with no theoretical import. The unaccusativity of the I is syntactic and is not determined by the lexical semantic relation between verb and argument, whether derived or basic. The syntactic distinction between unaccusative and unergative verbs is unmotivated. Semantic differences between the two classes remain, but they are seen in their role of contributing to sentence interpretation rather than determining syntactic structure. Free mapping of arguments, constrained by compatibility requirements, accounts in a straightforward way for the verb-argument pairs seen in the I ; moreover, the question of the Case of the PVNP is resolved through the straightforward assignment of structural Case. The interpretation of the I derives from the meaning of the construction and the individual semantic contributions of each of its elements. The presentational meaning of the I , as well as the range of verbs that are the most frequent and most easily interpretable in the construction, are seen as a consequence of its unaccusative configuration. The compositional approach also allows for an explanation of nonlexical elements, such as tense, in the I .
REFERENCES Auger, Julie. 1994. Pronominal Clitics in Quebec Colloquial French: A Morphological Analysis. Ph.D. dissertation, University of Pennsylvania. Belletti, Adriana. 1988. "The case of unaccusatives". Linguistic Inquiry 19.134. Birner, Betty & Gregory Ward. 1998. Information Status and Noncanonical Word Order in English. Amsterdam & Philadelphia: John Benjamins.
UNACCUSATIVITY
103
Ghomeshi, Jila & Diane Massam. 1994. "Lexical/syntactic relations without projection". Linguistic Analysis 24.175-217. Goldberg, Adele. 1995. Constructions: A Construction Grammar Approach to Argument Structure. Chicago: University of Chicago Press. Grimshaw, Jane. 1990. Argument Structure. Cambridge, Mass.: MIT Press. Guéron, Jacqueline. 1980. "On the syntax and semantics of PP extraposition". Linguistic Inquiry 11.637-678. Hériau, Michel. 1980. Le verbe impersonnel en français moderne. Lille: Atelier de reproduction de thèses, Université de Lille III. Herschensohn, Julia. 1982. "The French presentational as a base generated structure". Studies in Language 6. 193-219. . 1996. Case Suspension and Binary Complement Structure in French. Amsterdam & Philadelphia: John Benjamins. Hoekstra, Teun & René Mulder. 1990. "Unergatives as copular verbs: locational and existential predication". The Linguistic Review 7.1-79. Hulk, Aafke. 1989. "La construction impersonnelle et la structure de la phrase". Recherches linguistiques 18.59-79. Labelle, Marie. 1992. "Change of state and valency". Journal of Linguistics 28.375-414. Levin, Beth & Malka Rappaport Hovav. 1995. Unaccusativity: At the SyntaxLexical Semantics Interface. Cambridge, Mass.: MIT Press. Perlmutter, David. 1978. "Impersonal passives and the Unaccusativity Hypothesis". Proceedings of the Fourth Annual Meeting of the Berkeley Linguistics Society, 157-189. University of California, Berkeley. Pollock, Jean-Yves. 1986. "Sur la syntaxe de en et le paramètre du sujet nul". La Grammaire modulaire ed. by M. Ronat and D. Couquaux, 211-246. Paris: Éditions de Minuit. Pustejovsky, James. 1995. The Generative Lexicon. Cambridge, Mass.: MIT Press. Saccon, Graziella. 1992. "VP-internal arguments and locative subjects". Proceedings of the North East Linguistic Society 22.383-397. University of Massachusetts, Amherst. Safir, Kenneth. 1985. Syntactic Chains. Cambridge: Cambridge University Press. Zribi-Hertz, Anne. 1982. "La construction 'se-moyen' du français et son statut dans le triangle: moyen-passif-réfléchi". Linguisticae Investigationes VI:2. 345-401.
SHIFTING THE DP PARAMETER A STUDY O F ANGLOPHONE F R E N C H L2ERS
RANDALL GESS JULIA HERSCHENSOHN University of Utah University of Washington
0.
Introduction There is only one human language, apart from the lexicon, and language acquisition is in essence a matter of determining lexical idiosyncrasies. [...] If substantive elements (verbs, nouns, etc.) are drawn from an invariant universal vocabulary, then only functional elements will be parameterized. (Chomsky 1995: 131)
The mastery of L2 functional categories is a topic of importance not only to acquisition research, but also to syntactic theory, for in minimalist terms functional category features are the motivating force of the syntax. Furthermore, the path of acquisition provides a testing ground for the minimalist idea that cross-linguistic variation is primarily lexical, not syntactic. Our paper examines parametric variation between French and English Determiner Phrase (DP) in terms of two contrasting approaches to L2 acquisition, the Full Functional Hypothesis (FFH) and the Structure Building Hypothesis (SBH). We first discuss the parametric variation; then we present the two approaches to L2A. Finally, using a cross-sectional study of anglophone French L2ers' acquisition of DP syntax, we argue that our empirical evidence supports FFH over its competitor. 1. DP parameter 1.1 French and English English DPs differ from French according to two morphological criteria—gender (an intrinsic feature of all nouns) and agreement (the percolation of gender/number features throughout the DP)—and two syntactic criteria—overt determiner and adjective placement. This section reviews the parametric differences and presents an account of adjective
RANDALL GESS & JULIA HERSCHENSOHN
106
placement based on noun-raising proposals that have been put forth in recent studies. French and English differ according to several characteristics of the DP, as demonstrated in (l)-(4). (1)
(2)
(3)
(4)
le [m,sg] livre [m,sg] the book "the English book" la [f,sg] livre [f,sg] the pound "the English pound" les [m,pl] livres [m,pl] the books "the white books" les [f,pl] lèvres [f,pl] the lips "the white lips"
anglais [m,sg] English anglaise [f,sg] English blancs [m,pl] white blanches [f,pl] white
French, as the other Romance languages, designates nouns as masculine or feminine, a gender feature that percolates through the entire DP.1 Number and gender are evident in French in the morphology of determiners and adjectives that agree in gender and number with the noun in the written forms. In the spoken language it is the determiner that reveals gender and number, since the noun does not. In minimalist terms, the features of gender and number are [+interpretable] (Chomsky 1995) since they are necessary for the interpretation of the sentence. They persist at spell-out to insure Full Interpretation. The determiner is almost always obligatory in French, another difference with English which has a null determiner in plural generic (5) and partitive (6).2
1
Most nouns, like those in (l)-(4), do not have a natural gender, and the gender of the noun is not signaled morphologically in French by a noun marker such as masculine —o versus feminine -a in languages such as Spanish or Italian. Aside from predictable derivational suffixes such as -tion (f) or -isme (m), gender in French is idiosyncratic and must be learned as part of the lexical information associated with a given noun. 2 In other referential uses the two languages often agree on choice of definite and indefinite articles (e.g. first mention is indefinite, presupposed existence definite).
SHIFTING THE DP PARAMETER
(5)
(6)
Les castors construisent the beavers build "Beavers build dams." Je bois du café. I drink some coffee "I drink coffee."
107
des digues. some dams
Summarizing, the two morphological differences are the French requirements of intrinsic gender and agreement, and the two syntactic differences are the French overt determiner requirement and the placement of adjectives, to which we now turn. 1.2 Noun raising This last difference between French and English is the placement of the attributive adjective: in English it precedes the noun and in French it follows, as is evident in (l)-(4). This systematic variation has led several scholars to propose a DP parameter distinguishing languages with respect to adjective placement and gender, number and determiner realization (e.g. Abney 1987; Mallen 1990, 1991; Picallo 1991; Delfitto & Schroten 1991; Bernstein 1991, 1993; Valois 1991; Authier 1992). These works argue that noun phrases are complements of a determiner, the head of a DP that contains, in addition to the NP, other functional and lexical projections.3 Longobardi (1994) casts the DP parameter into a minimalist framework by suggesting that it is strength of functional features that motivates movement or non-movement of N.4 What these proposals have in common is the idea that certain morpho-syntactic nominal features are related to functional categories projected between D and NP, and that the features are shared or checked by the N by head movement of the N through these nodes. Chomsky (1995) proposes that parametric variation is determined by strength of [-interpretable] features of functional categories. If a given feature is strong, it will force overt raising (in the syntax) of a term with that feature so that the strong feature may be checked off before spell-out. Strong 3
Between D and its complement NP, Picallo (1991) proposes a gender phrase necessary for languages that mark overt gender, while Delfitto & Schroten (1991) propose a Number Phrase necessary for checking singular / plural. Valois (1991) argues for a Case Phrase and Bernstein (1993) for a "Word Marker (WM)" Phrase in addition to Number. 4 He motivates the distinction in terms of a feature "R" (referential) that is strong in Romance and weak in Germanic. The feature [R] of the D position must be checked either overtly in Romance which requires raising, or covertly in Germanic that has LF raising of N. Ishikawa (1999) applies this analysis to Spanish.
RANDALL GESS & JULIA HERSCHENSOHN
108
[-interpretable] features that persist at PF cause a derivation to crash; weak features do not. Adapting elements of a noun-raising treatment of DP, we assume that a strong [-interpretable] nominal feature F N in a higher functional category forces overt raising of N in French, but not in English. We have simplified the functional projections to an Agr/Num Phrase between D and N for the purposes of exposition. In order to be interpretable at LF, in both English and French the noun needs to raise to Agr/Num to check [+interpretable] morphological features. In French the N raises in the syntax (7), whereas in English it raises at LF (indicated by curly brackets in (8)). (7)
(8)
le livre anglais [DP [D le] [ Agr/NumP [livre,]
[NP [AP
anglais] [N t¡ ]]]]
the English book LDP [D the] [ Agr/NumP
{booki} [NP [AP English] [N booki]]]] The strong nominal feature of Agr/Num forces overt raising of N in French to that node, where it checks off the strong feature. In English N raises covertly at LF (indicated by curly brackets) to check its interpretable features. This analysis accounts for the word order contrast for attributive adjectives between the two languages. 2. L2A Theory 2.1 The Structure Building Hypothesis (SBH) Two approaches to the role of functional categories and morphological inflection in the emerging L2 grammar are put forth in recent studies, which we refer to as the Structure Building Hypothesis (SBH) and the Full Functional Hypothesis (FFH). SBH holds that the early L2 grammar has initially only lexical projections linked to incomplete morphology. There are three important assumptions underlying this hypothesis (9). (9)
Assumptions of SBH a. functional categories in L2 are initially absent or underspecified (Minimal Trees, Vainikka & Young-Scholten 1996, 1998) b. their activation is directly linked to specification of morphology (underspecification, Eubank & Grace 1998) activation takes place in stages (Hawkins 1999)
SHIFTING THE DP PARAMETER
109
Vainikka & Young-Scholten (1996, 1998) use longitudinal data from adult L2ers learning German to argue that the initial L2 grammar is incomplete with respect to functional categories, but that native language lexical categories and headedness transfer. They propose that the early interlanguage grammar—which overtly lacks IP, CP, agreement, complementizers, verb raising and overt subjects—develops subsequent stages as in (10). (10) VP>FP>AgrP>CP FP is underspecified tense and AgrP is IP.5 Extending this approach to DP, the early adult L2 grammar would have an NP stage without determiner, movement or agreement and would build stages with the intermediate functional projections as in (11). (11) NP>Agr/NumP>DP Eubank & Grace, in a complementary vein, also propose a "defective" L2 initial grammar and relate syntactic movement to the specification of L2 functional categories. They use morphological accuracy as a litmus test for syntactic movement.6 SBH has been challenged on the grounds that it cannot account for L1 transfer effects or for the apparently partial acquisition of functional categories at intermediate levels (Schwartz & Sprouse 1996; Epstein et al. 1998). 2.2 The Full Functional Hypothesis The second approach, FFH, holds that the early L2 grammar may access all functional projections, which are not directly dependent on the mastery of morphological inflection. This hypothesis incorporates three crucial characteristics (12).
5
This kind of analysis is proposed for L1 acquisition of DP by Clahsen et al. (1994: 102) who argue that the early nominal tree does not contain agreement features or D, "the child can only use uninflected determiners and adjectives or default forms of these elements." (cf. Radford 1990). 6 Hawkins (1998, 1999) proposes a "modulated structure building" account of the data he collects concerning DP agreement in French by anglophone L2ers.
110
RANDALL GESS & JULIA HERSCHENSOHN
(12) Assumptions of FFH a. functional categories in L2 are initially available (Full Transfer / Full Access, Schwartz & Sprouse 1996) b. their activation is not linked to the specification of morphology (Missing Inflection, Lardière 1998) c. syntax is acquired before idiosyncratic lexicon (Constructionism, Herschensohn 2000) According to this model, "the L1 grammar [...] is the L2 initial state [which] will have to change in light of the target language input." (Schwartz & Sprouse 1996:41) The L2er has Full Access to all UG possibilities although they are not necessarily employed. Activation of functional categories is not dependent on acquisition of L2 morphology; rather, morphological errors are attributed in part to problems in mapping between syntax and morphology.7 Missing inflection (Lardière 1998) may be phonetically unrealizable in fossilized grammars with impoverished morphology, rather than representing a mismatching of agreement features in the syntax. Finally, emphasizing the minimalist claim that syntax is universal, the Constructionist hypothesis (Herschensohn 2000) argues that the most salient aspects of a given parameter—those easily observable from positive evidence such as noun-adjective word order—should be the most accessible. On the other hand, morpholexical differences such as intrinsic gender, should be less obvious and perhaps more difficult to acquire since cross-linguistic variation is primarily in the lexicon. Two recent papers concerning DP acquisition support FFH. Bruhn de Garavito & White (2000), who study francophone students of L2 Spanish, show that the subjects do not go through a stage of underspecification, but rather show L1 transfer in noun-raising, have some missing inflection, and demonstrate some weakness with respect to idiosyncratic lexical features. Parodi et al. (1997), who look at acquisition of German DP by speakers of Korean, Turkish, Italian and Spanish, also corroborate L1 transfer and missing inflection.
7
Lardière argues that such difficulties are related to processing and morphological spell-out on the part of the L2er. "It may be this mapping itself—in the morphology or PF component—which is imperfectly acquired, and from which the status of syntactic phrase structure might therefore not be reliably inferred." (Lardière 1998:2) •-
SHIFTING THE DP PARAMETER
111
3. Empirical evidence 3.1 The task In order to test the two hypotheses we gathered cross-sectional data to examine suppliance of correct forms in the use of determiner, adjective placement, agreement and intrinsic gender by 85 college students studying French language. The procedures used and the results are presented and then discussed in this section. We limit our study to instructed learners.8 To gather data, we administered a 30-item written DP production task to four levels of anglophone French L2ers at the University of Utah (UU) and the University of Washington (UW) shown in Table 1. Level (Univ)
number
comments
1010 (UU) 2010 (UU) 3040 (UU)
n=14 n=29
true beginners intermediate classroom based 3 rd year
3060(UU)
n=14 n=20
400 (UW)
n=7
>12 mos immersion (3 rd year) 6 mos immersion (4 th year)
Table 1 : Subjects of DP task The 1010 course is the regular first-semester course for true beginners.9 The results suggest a bi-modal separation of the first-semester group into 01010 (n=6), with no determiners, and 1010 (n=8), with determiners (see below for discussion). UU maintains two fifth-semester courses, one for learners whose acquisition has been almost exclusively classroom based (3040), and one for those who have spent two years or more in a target language environment (3060, principally for returned missionaries).10 The UW course was taken by students who had already completed all language courses, six of whom had 8 Studies such as White & Juffs (1998) and Schlyter (1998) find few differences between instructed and so-called naturalistic learners. See Chapter 7 of Herschensohn (2000) for a more detailed discussion. 9 First-semester language students at UU recording any prior experience are required to take a placement exam and if they do not place into a second-semester or higher course, they are directed to an accelerated first-semester section. 10 Results from the two courses were very similar (as a peek forward at Table 3 shows). Both groups have virtually identical accuracy rates for determiner suppliance, adjective placement and correct gender assignment, but the "immersion group" scores 86% to the instructed group's 9 1 % accuracy on adjective agreement. The small difference with respect to agreement is interesting, and perhaps a surprising result for "naturalistic learning", but is not relevant to our study.
112
RANDALL GESS & JULIA HERSCHENSOHN
spent 6 months in France.11 The tests were administered during the seventh and eighth weeks of instruction at UU and during the third week of instruction at UW. The task items were four sets of incomplete sentences consisting of a subject and verb and requiring a DP to make them complete (Appendix 1). The sets included 5 or 10 sentences with a prompt and DP items (nouns and adjectives). For example, the first prompt Jean boit 'John drinks' requires completion with a beverage and a temperature (e.g. a cold beer). For all four sets, subjects were provided with the lexical items to combine. The format of the first set is shown in (13). (13) Sample acquisition task sentences In these sentences, put together the information below. Jean boit "Jean drinks " Temperature: froid chaud frais tiède cold hot cool lukewarm Drinks: bière café eau vin beer coffee water wine The task was designed to provide only the basic lexical triggers, thus requiring the students to determine intrinsic gender and number, to furnish determiners and agreement markers and to place the attributive adjective correctly after the noun. The task is, as one reviewer notes, a simplistic, familiar classroom activity, one for which ample models are provided. One might then expect perfect performance, even by first-year learners who are taught determiners on Day One. However, as our results demonstrate, it is not the case that all levels of learners performed with equal accuracy.
11 This course on French syntax had not covered any determiner features or DP phenomena. During the first three weeks it treated the lexicon, phrase structure rules and selection.
SHIFTING THE DP PARAMETER
113
3.2 Results The predictions of the two L2 approaches are summarized in Table 2.
SBH Bare NP stage (no D) Linked morpho-syntax (NR // agr) Stages (NP>Agr/NumP>DP )
FFH D initially available Sep. morpho-syntax (NR > agr) Lexicon last (NR>agr> intrinsic gender)
Table 2: Predictions for DP task SBH presupposes that anglophone learners of French will initially show bare NP with prenominal adjective. The next stage, Agr/NumP, would, by analogy with the FP stage, show no determiner (D), some noun raising (NR), and little agreement (agr). The last stage, DP would have overt D, and almost perfect noun raising and agreement. FFH allows D from intial state on and assumes that acquisition of syntax is earlier and more complete than acquisition of agreement which is hindered by missing inflection. It assumes that idiosyncrasies of the lexicon such as intrinsic gender are mastered last. The four French errors examined are exemplified (and underscored) in (14). (14) Sample errors of DP task (from 1010) a. Missing determiner: Marie porte jean bleu. "Marie wears jean blue." b. Adj-Noun order: Jean boit une tiède bière. "Jean drinks a lukewarm beer." c. Inconsistent agreement: Marie préfère ses (p1) amis (p1) patient (sg). "Marie prefers her friends patient." d. Faulty lexicon: Marie porte des (p1) chaussettes (f, p1) violets (m, pi). "Marie wears some socks purple" ... une chemise blance. [correct form blanche] "... a blouse white"
114
RANDALL GESS & JULIA HERSCHENSOHN
The errors in (14a-c) are self-explanatory. In (14d), there is a lexical error in the apparent assignment of masculine gender to the feminine noun chaussettes, evidenced by the use of the masculine form of the adjective.12 In order to calculate accuracy, the number of errors was divided by the total number of contexts (30 x the number of students minus a few missing items) and subtracted from 100%. The results of the task are represented in Table 3, and the actual number of errors/contexts is given in Appendix 2.
01010 1010 2010 3040 3060 400
Pet (DP)
Adj (NR)
Agr (morph)
Lex
6% 77% 92% 100% 99% 100%
34% 91% 93% 99% 100% 100%
NA 76% 86% 91% 86% 98%
63% 65% 76% 87% 86% 93%
Table 3: Suppliance in obligatory context; mastery of morpho-syntax 3.3 Discussion The results favor FFH over SBH since there is little evidence for a bare NP stage, for linked morpho-syntax, or for staged development. The bare NP and Agr/NumP stages (with no D) are contradicted even at level 1 where the majority of students furnished determiners even in contexts that allowed null articles in English.13 Furthermore, the linking of NR and overt morphology is disproved by the marked contrast between the mastery of syntax (D, NR) and that of gender morphology. There is no codependent mastery of syntax and morphology. We look at the results in detail, using the criteria of Table 2. By isolating the six students in 1010 who used virtually no determiners, we established a subclass (01010, n=6) of the first year learners who might be argued to constitute the stage of bare NP, and in this case might furnish some support for SBH. They show only 6% suppliance of determiners but are better at adjective word order with 34% accuracy. They use the default form of the adjective at virtually 100%, an indication that they have no rule for 12 Note that the indefinite plural article des is not overtly marked for gender, so that it reveals nothing with respect to choice of intrinsic gender for the noun. This error is considered to be one of incomplete lexical knowledge, and not lack of agreement, since the subject shows evidence of the latter by marking plural number throughout the DP. 13 None of our subjects showed a consistent pattern of only omitting determiners where they are missing in English.
SHIFTING THE DP PARAMETER
115
agreement.14 Although it could be argued that the zero group presents a bare NP profile, the presence of some N raising implicates a functional category in their grammars. As for the null determiners, FFH could argue that they constitute a UG option. Furthermore, while FFH can explain the early lack of overt determiners by assuming that the functional categories either have null terminals or are not projected, SBH cannot explain the sudden appearance of functional categories without resorting to additional mechanisms in the grammar.15 FFH's assumption of null terminals or lack of projection can accommodate eventual acquisition of functional categories and intermediate partial acquisition. The remaining eight students at the 1010 level show 77% accuracy in suppliance of determiners (over twelve times greater than their peers), correct adjective order at 91%, and 65% accuracy on idiosyncratic lexical information. Five of this group show evidence of an agreement rule, with 76%o accuracy (see Appendix 2). At the second year, learners increase scores in all areas: determiners at 92%, adjective order at 93%, agreement at 86%, and lexical at 76%. The syntactic differences—D and NR—between the 0 level students and the other first and second year learners are essentially precipitous. By second year the L2ers have virtually mastered the obligatory determiner and adjective order. The third and fourth year students have almost 100% syntax (determiners and adjective order) and better than 85% agreement / lexical. The evidence clearly shows that learners master the syntax before the morphological rule, and the morphology before the lexical information; and second that they learn the rule-governed morphosyntax in a definitive manner fairly early on for the DP. At the first year level, after only a few weeks of instruction in French, learners have understood that there are syntactic and morphological differences between English and French. They are already furnishing overt determiners in French where there are null articles in English; first-year learners place the attributive adjective after the noun in French. Their agreement progresses from first year to fourth year, but the agreement rule is acquired gradually, not precipitously. It increases from an absence of an agreement rule (01010) to 76% (1010) to 86% (2010) to 88% (3040 & 3060) to 98% (400). Finally, lexical mastery shows the lowest slope of increase: from 64% (01010, 1010) to 76% (2010) to 86% (3040, 3060) to 14
Consequently, they furnished correct tokens for required masculine singular forms; these were not counted ("NA") since they demonstrated no knowledge of a gender rule. 15 Herschensohn (2000:92-94) summarizes several points that have been argued against Structure Building in L1A. See also Schwartz & Sprouse (1996).
116
RANDALL GESS & JULIA HERSCHENSOHN
93% (400). There is no indication that morphological mastery increases in tandem with syntactic mastery of the overt determiner and noun raising in French, whereas it is clear that the French syntactic values of overt determiner and noun-raising are mastered early in the acquisition process. Only FFH can account for these data. The claim that L2 syntax is more accessible than the L2 lexicon is borne out by the fact that noun-raising becomes automatic earlier than lexically idiosyncratic, intrinsic gender is acquired. Knowledge of intrinsic gender and irregular morphology must be gained in a progressive manner, whereas a straightforward syntactic parameter that is consistent for most lexical items is available for a seemingly precipitous shift. Such are the predictions of FFH, which disproves the hypotheses of bare projections, staged functional categories, and linked morphology / syntax. 4.
Conclusion The cross-sectional data from anglophone French L2ers mastering the DP parameter support FFH over SBH. The former holds that the L2 grammar may access all possible functional categories, which are not directly dependent on the mastery of morphological inflection. L2 morphological incompleteness is due to difficulties related to processing, morphological spell-out and the idiosyncratic nature of intrinsic gender assignment. Our results demonstrate that L2ers are able to master functional features and movement associated with them even when the L1 differs with respect to these properties. This treatment accounts for longitudinal development while providing a linguistically motivated and consistent account of the data.
Appendix 1: Language Acquisition Task Sentence Completion Task Please write out complete sentences using as the base the sentence beginning that is already provided (e.g. "Jean boit..." in the first group). Then complete the sentences by choosing an item and a descriptor (e.g. "drinks" and "temperatures" in the first group) from the two rows provided in bold. Choose one word from each row (in the order given), and use each item only once. Finally, you will need to add any necessary changes such as agreement, articles, etc. Here's an example in English: John is drinking an icy coke.
SHIFTING THE DP PARAMETER
117
In this first group of sentences put together the information in the two rows (temperature, drinks) to tell what Jean chooses to accompany his food. Temperature: froid, chaud, tiède, frais, chambré Drinks: bière, chocolat, café, eau, vin Jean boit Marie is careful in choosing her associates; they must have the right characteristics. Characteristics: courageux, sportif, sérieux, agréable, patient Associates: femmes, hommes, professeurs, parents, amis Marie préfère Jean likes to eat an international diet. Choose from nationality and food to describe it. Nationality: français, espagnol, belge, allemand, mexicain, chinois, italien, américain, canadien, suisse Food: tarte, fraises, frites, jambon, haricots, salade, glace, pomme, fromage, croissants Jean mange In this group of sentences describe the color of the clothes that Marie is wearing today. Colors: vert, blanc, rouge, bleu, noir, violet, orange, gris, rose, orange Clothes: veste, chemise, bottes, jupe, jean, chausettes, short, chaussures, chapeau, robe Marie porte
Appendix 2: Number of errors/contexts (tokens)16
01010 1010 2010 3040 3060 400
16
Pet (DP)
Adj (NR)
Agr (morph)
Lex
169/180 52/228 69/867 0/419 5/599 0/210
123/180 20/228 62/867 3/419 1/599 0/210
NA 33/139 98/717 39/419 83/599 5/210
66/180 81/228 211/867 54/419 85/599 14/210
The denominator (contexts) represents (30 items x the number of subjects) minus tokens missing on incomplete tests. In calculating the figures for Agreement, subjects who simply supplied the furnished masculine singular adjective form for all task items were not included since they demonstrated no conceptual knowledge of agreement. (The calculations on which our conclusions are based are therefore quite conservative.) The subjects that were counted for Agreement are n=5 for 1010 and n=24 for 2010.
118
RANDALL GESS & JULIA HERSCHENSOHN
REFERENCES Abney, Stephen. 1987. The English NP in its Sentential Aspect. Ph.D. Dissertation, MIT. Authier, J.-Marc. 1992. "Is French a null subject language in the DP?" Probus 4.1-16. Bernstein, Judy. 1991. "DPs in French and Walloon: Evidence for parametric variation in nominal head movement". Probus 3.101-126. . 1993. "The syntactic role of word markers in null nominal constuctions". Probus 5.5-38. Bruhn de Garavito, Joyce & Lydia White. 2000. "L2 acquistion of Spanish DPs: The status of grammatical features". Proceedings of the 24th Annual Boston University Conference on Language Development ed. by S. Catherine Howell, Sara A. Fish & Thea Keith-Lucas, 164-175. Somerville, Mass.: Cascadilla Press. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. Clahsen, Harald, Sonja Eisenbeiss & Anne Vainikka. 1994. 'The seeds of structure: A syntactic analysis of the acquisition of case marking". Language Acquisition Studies in Generative Grammar ed. by Teun Hoekstra & Bonnie Schwartz, 85-118. Amsterdam and Philadelphia: John Benjamins. Delfitto, Denis & Jan Schroten. 1991. "Bare plurals and the number affix in DP".Probus 3.155-185. Epstein, Samuel D., Suzanne Flynn & Gita Martohardjono. 1998. "The strong continuity hypothesis: Some evidence concerning functional categories in adult L2 acquisition". The Generative Study of Second Language Acquisition ed. by Suzanne Flynn, Gita Martohardjono & Wayne 'Neil, 61-77. Mahwah, NJ: L. Erlbaum. Eubank, Lynn & Grace, Sabine. T. 1998. "V-to-I and inflection in nonnative grammars". Morphology and its Lnterfaces in Second Language Knowledge ed. by Maria-Luise Beck, 69-88. Amsterdam & Philadelphia: John Benjamins. Hawkins, Roger. 1998. "Explaining the difficulty of French gender attribution for speakers of English". Paper presented at the EUROSLA Conference. . 1999. Second Language Syntax: A generative introduction. Ms., University of Essex, Essex.
SHIFTING THE DP PARAMETER
119
Herschensohn, Julia. 2000. The Second Time Around: Minimalism and L2 acquisition. Amsterdam & Philadelphia: John Benjamins. Ishikawa, Masataka. 1999. "Morphological strength and syntactic change". Linguistic Inquiry 30.301-310. Lardière, Donna. 1998. "Case and tense in the fossilized steady state". Second Language Research 14.1-26. Longobardi, Giuseppe. 1994. "Reference and proper names: A theory of N movement in syntax and Logical Form". Linguistic Inquiry 25.609-665. Mallén, Enrique. 1990. "Clitic movement inside noun phrases". Studia Linguistica 44.1-29. . 1991. "Noun phrase structure, clitic doubling and extraction". Linguistische Berichte 134.276-309. Parodi, Teresa, Bonnie Schwartz & Harald Clahsen. 1997. "On the L2 acquisition of the morpho-syntax of German nominais". Essex Research Reports in Linguistics 15.1-43. Picallo, M. Carme. 1991. "Nomináis and nominalizations in Catalan". Probus 3.279-316. Radford, Andrew. 1990. Syntactic Theory and the Acquisiton of English Syntax. Oxford: Blackwell. Schlyter, Suzanne. 1998. Acquisition du français parlé - une comparaison entre apprenants formels et informels. Paper presented at the Colloque sur le français parlé. Schwartz, Bonnie & Rex Sprouse. 1996. "L2 cognitive states and the Full Transfer / Full Access model". Second Language Research 12.40-72. Vainikka, Anne & Martha Young-Scholten. 1996. "Gradual development of L2 phrase structure". Second Language Research 12.7-39. 1 1998. "Morphosyntactic triggers in adult SLA". Morphology and its Interfaces in Second Language Knowledge ed. by Maria-Luise Beck, 89114. Amsterdam & Philadelphia: John Benjamins. Valois, Daniel. 1991. "The internal syntax of DP and adjective placement in French and English". NELS 21, 367-382. Amherst: GLSA. White, Lydia & Alan Juffs. 1998. "Constraints on Wh-movement in two different contexts of normative language acquisition: Competence and processing". The Generative Study of Second Language Acquisition ed. by Suzanne Flynn, Gita Martohardjono & Wayne O'Neil, 111-129. Mahwah, NJ: L. Erlbaum.
CONSTRAINT DEMOTION AND NULL-SUBJECTS IN SPANISH L2 ACQUISITION* LARRY LAFOND
University of South Carolina
RACHEL HAYES
University of Arizona
RAKESH BHATT
University of Illinois
0.
Introduction A fundamental concern within second language acquisition (L2A) research has been the description and characterization of the learning mechanism speakers use to resolve conflicts between what they know (their first language system) and what they are learning (the target language system). The second language (L2) learning process can thus be conceived of as a process of conflict and its resolution, evidenced in developmental stages. One well-observed example of such a conflict is the difference between languages such as English or French, that prohibit null-subjects in tensed clauses, and languages such as Spanish or Italian, that in certain discoursal contexts require null-subjects in such clauses. For example, if one conversational participant asks the question 'What did you see?' the response in (la) is fully grammatical in Spanish, but its English counterpart (lb) is not. (1)
a.
b.
Vi la casa de Maria. saw-lsg the house of Maria "I saw Maria's house." * saw the house of Maria.
A natural question to ask then is how and why learners follow the developmental path that they do. Since the problem of language learning may *Many thanks to Ramona Lagos, Maria Mabrey, Carolyn Hansen, and their students who assisted us with the collection of Spanish data. Thanks also to the audiences of the 19th Annual SLRF and the 30th Annual LSRL, as well as to Eric Holt, Mike Hammond, Barbara Hancin-Bhatt, Juana Liceras, Ana Teresa Pérez-Leroux, and Jacqueline Toribio for their comments, suggestions, and help in various ways. Finally, we would like to thank two anonymous reviewers for their helpful comments and suggestions. The usual disclaimers, of course, apply.
122
LARRY LAFOND, RACHEL HAYES & RAKESH BHATT
be divided into issues of parsing and grammar learning (Tesar & Smolensky 1998), a theory of grammar is needed that provides a sufficient characterization of learners' successive attempts to understand and produce target language (TL) structures. We will argue in this paper that Optimality Theory (Prince & Smolensky 1993) economically characterizes the developmental stages in the acquisition of null-subjects in L2 Spanish as a function of the learning procedure - the Constraint Demotion Algorithm (Tesar & Smolensky 1998). In so doing, we will also further a line of research that has demonstrated the need to consider semantic and discoursal interfaces with syntax in the second language acquisition of null-subjects (see White 1985, Liceras 1988, Pérez-Leroux & Glass 1997, Liceras & Díaz 1999). 1. Theoretical preliminaries 1.1 Optimality Optimality Theory ( ) is essentially a theory of constraint interaction and how grammars are defined by constraint hierarchies (McCarthy 1995). Universal Grammar (UG) in provides a finite set of potentially conflicting violable constraints on structural well-formedness. Thus, the theory departs from standard generative frameworks in that instead of using categorical constraints to express empirical generalizations, uses violable constraints. These constraints are violable in just those contexts in which they conflict with a higher ranked constraint, and languages differ from each other in terms of how each ranks the set of constraints. The core ideas of can be summed up in the following way (Grimshaw 1997): (a) constraints can be violated; (b) constraints are ranked; and (c) the optimal form is grammatical. The grammatical system essentially works in the following way: An input, e.g., argument structure with operator features assigned, is fed into a generator (GEN). This generates, using an X'-theoretic system, all possible structural descriptions of the input, called candidates. Candidates are then evaluated by the harmony evaluator (EVAL), which contains the languagespecific constraint hierarchy and determines an optimal grammatical output the one with the least serious constraint violations. Of interest in this paper is to explore how learner grammars resolve conflicts that arise when discoursal requirements of their L2 Spanish require that a subject coreferent with the topic be null, whereas L1 English syntactic requirements override the discourse grammar requirements forcing the nonnull option. Such cases thus present linguistic scenarios where the requirements of discourse and syntax are clearly in conflict with each other: a
NULL-SUBJECTS IN SPANISH L2 ACQUISITION
linguistic puzzle that, we believe, finds a natural account within an of constraint interaction.
123 view
1.2 Full transfer/full access Assuming the Full Transfer/Full Access model (FT-FA; Schwartz & Sprouse 1996), the initial state of L2 acquisition is a full instantiation of the first language (L1) constraint hierarchy. Subsequent L2 input feeds the language learning mechanism, which in turn causes the interlanguage (IL) grammar to restructure in order to conform to the L2 grammar (Tesar & Smolensky 1998). Thus, subsequent acquisition is guided by UG (full access). Native speakers of English learning Spanish must in the process of their L2 acquisition reorganize their initial (English) constraint hierarchies to conform to the hierarchies of the L2 (Spanish); specifically, they must prioritize discourse constraints over syntax constraints. The mechanism for restructuring is described in the next section. 1.3 The learning theory Assuming an unordered set of universal constraints provided by UG, Tesar & Smolensky (1998) propose that OT-grammar learning takes place via successive applications of the Constraint Demotion Algorithm (CDA), which detects differences between the developing and the input/ TL grammar and subsequently rearranges, by demoting, constraints in the learner's grammar in order to match more closely the grammar of the input. Extending this general principle to L2 acquisition, we claim a similar learning procedure implicated in L2 learning, with an important caveat: The initial state in L2 acquisition is not unordered; in fact, following the logic of the FT-FA model, we claim that it is a fully hierarchized representation of the L1 grammar. Given this framework of assumptions, we argue that L2 grammatical development is modeled by the CDA, which works on mismatches between the grammatical properties of the L2 input and the internalized system. The restructuring of the internalized system to accommodate L2 input yields a developmental path that can be observed in terms of new configurations of constraint rankings. There are two specific mechanisms invested in this model: Robust Interpretive Parsing (RIP) and Constraint Demotion. RIP enables the learner's grammar to assign an underlying representation to the linguistic input even though the learner's grammar is not yet able to accommodate the input. Constraint Demotion determines the particular constraint-hierarchical differences responsible for the inability to assign a
124
LARRY LAFOND, RACHEL HAYES & RAKESH BHATT
representation to the input and guides restructuring of the IL grammar. We discuss the details of this OT-learning procedure in Section 6.1. 2. Null subjects in The observation that discourse topics trigger the use of null subjects in some null-subject languages led Grimshaw & Samek-Lodovici (1995) to propose a set of constraints, the interactions of which account for the missing subject phenomena in Italian, a language which exhibits similar missing subject patterns to Spanish. Their analysis of topic-connected subjects in Italian and English makes use of the generalization that English violates discourse constraints in order to maintain the integrity of the syntactic structure, while Italian (and Spanish) prioritize discourse over syntax in the case of subjects which are coreferent with the discourse topic. They propose the discourse constraint DROPTOPIC (requiring coreferential topic-connected subjects be covert), the syntax constraint SUBJECT (requiring clauses to have overt subjects), and the Faithfulness constraint PARSE (which requires the parsing of input constituents). When the subject of a sentence is topic-connected, the DROPTOPIC constraint is in conflict with both PARSE and SUBJECT. In Italian, DROPTOPIC outranks PARSE and SUBJECT in order for constructions with null topic subjects to be optimal, as demonstrated in Tableau 1. DROPTOPIC
a. b.
0 ha cantato (null) lui ha cantato ha cantato lui
PARSE
* **!
SUBJECT
*
*! *! * Tableau 1: Input:
tense = present perfect d.
PARSE must dominate SUBJECT because the winning candidate (a) has a violation of subject while a losing candidate (b) satisfies SUBJECT. Reversing the order between these two constraints will yield the ungrammatical, radically null parse, as optimal. And DROPTOPIC necessarily ranks above the other two, since losing candidates violate DROPTOPIC but do not violate SUBJECT. PARSE. Thus the hierarchy for Italian is: DROPTOPIC » PARSE » English, on the other hand, has a different constraint hierarchy for these constraints, as English does not allow topics to be dropped in the way that Spanish and Italian do. In English, SUBJECT must outrank DROPTOPIC to ensure that candidates in which the subject has been dropped turn out to be sub-optimal. Also, PARSE must outrank SUBJECT in order to prevent null
NULL-SUBJECTS IN SPANISH L2 ACQUISITION
125
candidates from (vacuously) satisfying SUBJECT. These interactions are demonstrated in Tableau 2.
a. b.
0 has sung (null) he has sung has sung he
PARSE
SUBJECT
*!
*
DROPTOPIC
*i*
* *! * Tableau 2: Input: <sing (x), x = topic, x = he> tense = present perfect '
d.
The optimal candidate in English is thus allowed to violate DROPTOPIC; the non-optimal candidates are eliminated for PARSE and SUBJECT violations. 3. Research questions and hypotheses We have argued thus far that differences between Spanish and English in regards to null-subjects are best understood by the relative rankings of discoursal and syntactic constraints. Therefore, we query whether the developmental path our learners take provides evidence for this competition between constraints. Will learners be able to correctly identify unacceptable possibilities in the TL grammar and will their developmental path follow the path predicted by the CDA? We propose two hypotheses, one corresponding to negative transfer (2a) and the other to positive transfer (2b), as plausible answers to the questions: (2)
a.
Intermediate and advanced levels will outperform beginners regarding the frequency of errors for topic-connected subjects.
b.
There will be no significant differences in the frequency of errors for non-topics among proficiency groups.
4. Methodology 4.1 Subjects Ninety-eight subjects, divided into five levels based on proficiency, participated in this study. Eighty-one of these were students in beginning (n=20), intermediate (n=18), and advanced (n=43) levels of Spanish study at the University of South Carolina. A near-native group (n=10) consisted entirely of graduate students, some employed as Spanish teaching assistants and some involved in an international business program at the university, all of whom had extensive language experience in Spanish-speaking countries. A control group of native speakers of Spanish (n=7) was also used.
126
LARRY LAFOND, RACHEL HAYES & RAKESH BHATT
4.2 Method Data was collected through the use of a written instrument designed to assess judgments regarding when native speakers would choose to use null subjects in conversation. All subjects read 40 incomplete constructed dialogues between two interlocutors in Spanish. Each dialogue consisted of three exchanges, with the first half of the final exchange being a question or statement that the learner was asked to respond to by selecting the most native-like of two provided responses. Subjects were given thirty minutessufficient time to complete the task. For all target items, the semantic content of each of the two choices was identical. Of the two choices, one was appropriate to the information structure and syntax of the dialogue, while the other violated the requirements of information structure or syntax. In some dialogues, the discourse context provided a clear topic choice, in which case a null-subject was the target choice (3). In other dialogues, the discourse context selected the non-topic choice, requiring an overt subject (4). The target answers for each of these items are presented here in boldface. (3)
A: B: A: B:
Hola, Ben. ¿Cómo estás? Bien, ¿y tú? Bien. ¿Cómo está tu esposa? Bien está. Ha estado muy ocupada en su trabajo. ¿Qué es lo que hace? 1) Trabaja en un banco. 2) Ella trabaja en un banco.
"Hi, Ben. How are you?" "Fine, and you?" "Fine. How is your wife?" "She's good. She's been very busy at work." "What does she do?" "Works at a bank." "She works at a bank."
A: B: A:
Hola John. Hola Ana ¿Te gustaría almorzar conmigo?
B:
Sí, me gustaría. ¿Puede venir Beth también? Seguro, ¿a qué hora quieres ir?
"Hi, John." "Hi, Ann." "Would you like to eat lunch with me?" "Yes, I would. Can Beth come, too?" "Sure, when do you want to go?" "Is in class now. Is 12:30 ok?" "Beth is in class now. Is 12:30 ok?"
A: B: (4)
A: B:
1) Bueno, está en clase ahora. ¿Está bien a las 12:30? 2) Bueno, Beth está en clase ahora. ¿Está bien a las 12:30?
5. Results There were two possible types of errors in this exercise: A topic error occurred when a subject inappropriately drops a non-topic subject (underuse of null subjects), and a non-topic error occurred when a subject
NULL-SUBJECTS IN SPANISH L2 ACQUISITION
127
inappropriately maintained a topic subject (overuse of overt pronouns). The charts and graphs below provide the results of the experiment.
Figure 1: Error patterns across proficiency levels
Group Beginners Intermediates Advanced Near Native Native Control
N 20 18 43 10 7
Mean - Total Errors 3.90 2.67 2.37 .111 .029
Standard Deviation .174 .153 .143 .033 .049
Table 1: Total Errors for topic- and non-topic-connected subjects (out of 10)
Group Beginners Intermediates Advanced Near Native Native Control
N 20 18 43 10 7
Mean - Topic Errors 1.7 .78 .67 .33 .29
Standard Deviation .108 .122 .071 .050 .049
Table 2: Errors for topic-connected subjects (out of 5)
128
LARRY LAFOND, RACHEL HAYES & RAKESH BHATT
Group Beginners Intermediates Advanced Near-Native Native Control
N 20 18 43 10 7
Mean - Non-topic 2.2 1.89 1.70 .78 0.000
Standard Deviation .136 .123 .130 .044 0.000
Table 3: Errors for non-topic-connected subjects (out of 5) We conducted an analysis of variance over the number of topic, nontopic, and total subject errors per group. The ANOVA obtained statistically significant main effects across groups for topic error F(4,92)=6.532, p=.0001, non-topic error, F(4,92)=5.687, p=0004, and total error F(4,92)=l1.369, p=.0001. A post-hoc Scheffe for topic errors (in Table 2) indicated that the difference between beginner and all other groups was statistically significant. For non-topic errors (in Table 3), there were statistically significant differences between beginners and control (p=.003), intermediates and control (p=.019) and advanced and control (p=.022). However, there was no significant difference between near-native and control groups. For total errors, there was no statistically significant difference between beginners and intermediates, though there was between beginners and all other groups. Total errors (in Table 1) also revealed statistically significant differences between all groups and the control group, with the exception of near-native and control, for which there was no significant difference. 5.1 Summary of Results The results indicate that learners initially do not drop topics as required by the L2 grammar. However, learners begin to use null-subjects early, without discriminating discourse context, initially resulting in overgeneration of null-subjects. As for subject retention in non-topic contexts, learners converge upon the TL grammar only in the advanced stages of acquisition. The gradual development evident in these results parallel learning patterns observed in Pérez-Leroux & Glass (1999). 6.
Discussion The results indicate that the learners show a clear developmental pattern in the acquisition of the correct ranking for the constraint DROPTOPIC, namely, that the subjects that are topic-connected are necessarily dropped. The data also show that the error rates for subjects that are not topicconnected show the most improvement between the intermediate and the advanced intermediate levels. We believe the initial increase in error rate for
NULL-SUBJECTS IN SPANISH L2 ACQUISITION
129
the dropping of non-topic connected subjects between the beginning and intermediate level is a function of the overgeneralization of the DROPTOPIC constraint; that is, while learners increasingly omit subjects which must be omitted, in the intermediate stages they overgeneralize to also omit subjects which must be maintained. In section 1.3, we discussed how the resolution of conflicting constraints of competing grammatical systems might be described through a learning mechanism of Constraint Demotion (CD). Using the L1 ranking as the initial hierarchy, CD demotes sub-optimal constraints as far down as necessary to converge on the TL ranking. The CD process may account for the data in our study.1 6.1 The learning algorithm In this section we follow Tesar & Smolensky (1998) and show how the Constraint Demotion Algorithm accounts for the developmental data found in the present study. Consider the three constraints given in Table 4 (Grimshaw & Samek-Lodovici 1995). Assume the L1 English constraint hierarchy in (5) and the TL Spanish grammar constraint ranking in (6): PARSE
Parse input constituents
SUBJECT
The highest A-specifier in an extended projection must be filled (Grimshaw 1997) Leave arguments coreferent with the topic structurally unrealized
DROPTOPIC
Failed by unparsed elements in the input Failed by clauses without a subject in the canonical position Failed by overt constituents which are coreferential with the topic
Table 4: Constraints on topics and subjects (5)
English (Initial State) ranking:
PARSE
(6)
Spanish (Target) ranking:
DROP TOPIC »
»
SUBJECT
»
DROP TOPIC
PARSE »
SUBJECT
Given L2 Spanish input such as (7); (7)
0 Trabaja en un banco work-3sg in a bank "He/She works in a bank."
GEN produces the candidates in (8): 1 One reviewer noted, and we agree, that the CD algorithm captures the gradual development in learners' grammars in a way that standard views of parameter resetting can not.
130
LARRY LAFOND, RACHEL HAYES & RAKESH BHATT
(8)
GEN: (a) (b)
[IP He[V? ...] [IP ø [VP ...]
These competing structures differ in the constraints they violate. The initial learner-grammar chooses candidate (a) as the optimal candidate when the subject is a topic (Tableau 3): Subject = topic a. [IP He[VP ...] b. [IP ø[VP ...]
PARSE
SUBJECT
DROPTOPIC
* *! * Tableau 3: English - the subject is a topic
The initial learner-grammar also chooses candidate (a) when the subject is a non-topic (Tableau 4): Subject = non-topic . [1 H [ ...] b. [IP ø[VP ...]
PARSE
SUBJECT
DROPTOPIC
*! * Tableau 4: English - the subject is not a topic
Unlike Spanish, the constraint ranking for English results in an invariant winner between topics and non-topics, the highest ranked Parse and Subject constraints are fatally violated regardless of whether a topic or non-topic is dropped. To arrive at the correct target grammar, the parse (b) in Tableau 3 must become optimal. This necessitates an error-driven restructuring of the learner's grammar, as detailed in (9) below: (9) Error-Driven Constraint Demotion Algorithm (EDCD) Given: hierarchy H and a set PositiveData of grammatical structural descriptions. For each description winner in PositiveData: set loser to be the optimal description assigned by H to I, the underlying form of winner. If loser is identical to winner, keep H; Else: apply Mark Cancellation, getting (marks' (loser), marks' (winner)); apply Constraint Demotion to (marks' (loser), marks' (winner)) and H; and adopt the new hierarchy resulting from the demotion as the current hierarchy. Tableau 5 shows the first loser/winner pair. Mark cancellation is applied to the pair of marks list:
NULL-SUBJECTS IN SPANISH L2 ACQUISITION
Loser/winner pairs [IP ø[VP ...] < [IP He [VP ...]
marks' (loser) DROPTOPIC
131
marks' (winner) SUBJECT
Tableau 5: Mark-data pair Given the logic of the learning algorithm (9), the winner-violated constraint SUBJECT is demoted below the loser-violated constraint DROPTOPIC:
SUBJECT
DROPTOPIC
*
a. [IP#e[VP ...] b. [IP0[VP ...]
*! Tableau 6: Demotion of the constraint
SUBJECT
The output of the first application of the recursive CDA results in an interlanguage (IL) hierarchy where PARSE » DROPTOPIC » SUBJECT. However this IL constraint hierarchy still chooses a target suboptimal as its optimal output, as shown in Tableau 7.2 Subject = topic a. [IP He[VP ...] b. [IP ø[VP ...]
PARSE
DROPTOPIC
SUBJECT
*
*! Tableau 7: L2 (intermediate level)
PARSE
»
DROP TOPIC
* »
SUBJECT
With further input, this stage of the IL grammar will once again require reranking, because the learning algorithm applies as long as there remains a mismatch between loser and winner marks. Tableau 8 shows the mark-data pair for loser/winner pair. Loser/winner pairs [IP ø[VP ...] < [1 H [
...]
marks' (loser) DROPTOPIC
marks' (winner) PARSE
Tableau 8: Mark-data pair
2
In this intermediate grammar, the (b) sub-optimal candidate now violates the first and third-ranked constraints rather than the first and second ranked constraints (see Tableau 3). If fewer serious constraint violations translate into more grammatical sequences, then we can account via this first demotion of SUBJECT below DROPTOPIC for the learner's improved performance at the intermediate stage.
132
LARRY LAFOND, RACHEL HAYES & RAKESH BHATT
Once again, following mark cancellation applies vacuously in Tableau 8, once again triggering constraint demotion, as shown in Tableau 9. The marks' winner PARSE is demoted below the marks' loser DROPTOPIC, resulting in a new constraint (target-like) hierarchy where DROP TOPIC
»
» SUBJECT. This constraint hierarchy now correctly chooses the optimal candidate, (b), the grammatical option for topic-connected subjects in L2 Spanish, as shown in Tableau 10. PARSE
Subject = topic a. [IP John [VP ...] b. [IP ø[VP ...]
PARSE
* *!
Tableau 9: L2 Interlanguage: Subject = topic a. [IP John [VP ...] b. [VP ø [VP ...]
DROPTOPIC
PARSE
»
DROP TOPIC
DROPTOPIC
Tableau 10: L2 Steady State:
»
PARSE
SUBJECT SUBJECT
*! * DROP TOPIC
»
PARSE
* »
SUBJECT
Since this ranking converges on the TL, the learning algorithm halts. Notice that the final restructured constraint ranking correctly reflects the observation that topics will be dropped; non-topics will also necessarily be retained by this ranking. Therefore, the Constraint Demotion Algorithm accounts for all stages of data manifested by our L2 learners of Spanish. 6.2 Conceptualizing Learning What do the results here suggest more generally regarding the syntaxdiscourse interface in the L2 acquisition of Spanish and Italian? Is there evidence in other structures that would support a shift in constraint rankings along the same syntax-discourse lines? We propose a model to schematize the relative prominence of syntactic versus discourse requirements in developing learner-grammars (Figure 2). At first (as in the learners' L1 English), syntactic requirements have priority over discourse requirements, and constraint "c" strictly dominates constraint "p". In the first IL stage, that dominance relationship is weakened to a tendency for "c" to outrank "p". Although the opposite ranking is also possible at this point, the single ">" indicates that the relationship is still one
NULL-SUBJECTS IN SPANISH L2 ACQUISITION
133
Figure 2: Developmental path of Spanish L2A by native speakers of English (Adapted from Bhatt 1999)
of tendency; that is, "c" tends to outrank "p" in this stage. In the second IL stage, "c" and "p" are unranked relative to each other, showing an apparent optionality with respect to the ranking of these two constraints. This optionality is realized in the productions of the learners as a lack of tendency towards one ranking or the other. In the third IL stage, the "p" tends to outrank constraint "c", but it is not yet a case of strict domination. And finally, in the final (target Spanish) state, there is again a strict domination relationship, with "p" dominating "c". 6.3 Residual issues Simply stated, L2 learners of Spanish have four logical options for dealing with TL outputs. They may: (a) drop topics, b) retain topics, c) retain non-topics, or d) drop non-topics. The first three of these options have natural explanations as realizations of either the target or L1 grammars of the learner. Learners who are dropping topics will be approximating the TL grammar and
134
LARRY LAFOND, RACHEL HAYES & RAKESH BHATT
learners who are retaining topics will be demonstrating transfer effects from their L1 English. Learners would be expected to retain non-topics in either grammatical system, so the retention of non-topics is not decisive evidence for either the TL or L1 grammars. But the deletion of non-topics suggests that an interaction between conflicting constraints is at work.
Topics Non-topic
Drop Spanish
Retain English English / Spanish
* Table 5: Logical possibilities in IL grammars
The analysis of null subjects that we have adopted cannot directly account for the fact that learners in our study dropped non-topic subjects. It is, however, conceivable that in the process of learning a language where topic subjects are dropped from an L1 where topic status has an impact on the syntax, learners are simultaneously learning what exactly a topic is and that if a subject is a topic, it needs to be dropped. In other words, it is possible that learners initially overgeneralize topic to mean subject, and when they learn that topics need to be dropped in certain cases, they simply drop all subjects. Further research is needed to investigate this issue (Cf. also Díaz-Rodríguez & Liceras 1990, Glass & Pérez-Leroux 1998). 7.
Conclusions In this study, we have extended previous research by studying the interaction of syntactic properties and discourse properties within a major component of information structure (Topic) as L1 English speakers learn L2 Spanish. Additionally, this work contributes to a fuller understanding of L2 acquisition by providing new insight into an insufficiently examined interface between discoursal and syntactic components of grammar. A primary goal in interpreting these results was to reveal a developmental path and determine whether a recursive constraint demotion learning algorithm can account for the path. We see the variant outputs of IL grammars as differences in their rankings of several syntactic and discoursal constraints. These results find a natural interpretation within an framework, and provide empirical evidence in support of the CDA of Tesar & Smolensky (1998). Finally, we propose a model of the IL stages exhibited by our learners that demonstrates the characteristics of strict domination, hierarchical tendencies, and optionality in the development of an L2 grammar.
LEARNING NULL-SUBJECTS
135
REFERENCES Bhatt, Rakesh. 1999. "Second Language Learning at the Syntax-Discourse Interface". Ms., University of South Carolina. Díaz-Rodríguez, Lourdes & Juana M. Liceras. 1990. "Formulación de Parámetros y Adquisición de Lenguas Extranjeras". V Congreso de Lenguajes Naturales y Lenguajes Formales ed. by . M. Vide, 465-480. Barcelona: Facultat de Filología, Universitat de Barcelona. Glass, William R. & Ana Teresa Pérez-Leroux. 1998. "A Parametric Interpretation of Learners' Errors: The Acquisition of Spanish Null Subjects". Estudios de Lingüística Aplicada 16.27-46. Grimshaw, Jane. 1997. "Projection, Heads, and Optimality". Linguistic Inquiry 28.373-422. & Vieri Samek-Lodovici. 1995. "Optimal Subjects". University of Massachusetts Occasional Papers in Linguistics 18.589-606. Liceras, Juana. 1988. "Syntax and Stylistics: More on the Pro-Drop Parameter". Learnability and Second Languages ed. by James Pankhurst, Michael Sharwood Smith & Paul VanBuren, 71-93. Dordrecht: Foris. & Lourdes Díaz. 1999. "Topic-Drop Versus Pro-Drop: Null-Subjects and Pronominal Subjects in the Spanish L2 of Chinese, English, French, German and Japanese Speakers". Second Language Research 15.1-40. McCarthy, John. 1995. "Some Misconceptions about Optimality Theory". Ms., University of Massachusetts, Amherst. Pérez-Leroux, Ana Teresa & William R. Glass. 1997. "OPC Effects on the L2 Acquisition of Spanish". Contemporary Perspectives on the Acquisition of Spanish, Vol 1: Developing Grammars ed. by Ana Teresa Perez-Leroux & William Glass, 149-165. Somerville, Mass.: Casadilla Press. Pérez-Leroux, Ana Teresa & William R. Glass. 1999. "Null Anaphora in Spanish Second Language Acquisition: Probabilistic Versus Generative Approaches". Second Language Research 15.220-229. Prince, Alan & Paul Smolensky. 1993. "Optimality: Constraint Interaction in Generative Grammar". Ms., Rutgers University & University of Colorado. Schwartz, Bonnie & Rex Sprouse. 1996. "L2 Cognitive States and the Full Transfer/Full Access Model". Second Language Research 12.40-72. Tesar, Bruce & Paul Smolensky. 1998. "Learnability in Optimality Theory". Linguistic Inquiry 29.229-268. White, Lydia. 1985. "The 'Pro-Drop' Parameter in Adult Second Language Acquisition". Language Learning 36.227-230.
TEMPORAL LOCATION OF EVENTS AND THE DISTRIBUTION OF THE ROMANCE COUNTERPARTS OF SINCE-ADVERBIALS* TELMO MÓIA Faculdade de Letras da Universidade de Lisboa
0.
Introduction The present paper focuses on linguistic data showing that quantification over events - in particular, associated with a summation operation - may act as a constraint on the distribution of some non-punctual temporal adverbials. This data involves mainly four Romance counterparts of English since-adverbials, namely those headed by Portuguese desde, Spanish desde, Italian da, and French depuis. For the sake of simplicity, throughout the text I will only supply examples with Portuguese desde-phrases, as its Romance equivalents exhibit a similar behaviour with respect to the issues at stake. It must be stressed that only the use of the expressions in question as 'inclusive adverbials' is relevant here. Such use is observable whenever the relevant eventuality described in the matrix clause is located somewhere within the time span associated with the adverbial, and not necessarily all through that span. For instance, in a sentence like John has visited Paris since 1980, the temporal adverbial can only be 'inclusive', as it locates the visit somewhere between 1980 and the moment of speech, but in a sentence like Paulo has been living in Paris since 1980, it is 'durative', as the state of Paulo living in Paris is said to hold all through the location time (see e.g. Vlach 1993, for this terminology). I will start by presenting - in section 1 - some linguistic evidence that the distribution of Portuguese desde-phrases is much more restricted than that of
* The core content of this text was presented at the Conference on (Preferably) Non-Lexical Semantics, Université de Paris VII, June 1996 (of which no Proceedings were published) and is part of my doctoral dissertation Identifying and Computing Temporal Locating Adverbials with a Particular Focus on Portuguese and English (presented to the University of Lisbon in February 2000). I thank João Peres, Hans Kamp and Rainer Bäuerle for insightful comments on previous versions of this text.
138
TELMO MÓIA
English sinc -phrases (according to the descriptions of since known from the literature). In fact, desde-phrases - and their Romance counterparts - seem incompatible with descriptions of events, except in some contexts that will be briefly described (in section 2), and whose unifying property appears to be that they involve, directly or indirectly, some form of 'event-quantification'. Roughly, the basic idea is that inclusive location by desde-adverbials is restricted to the 'supremum' of the set of sub-events which, on the one hand, are of the type described in the matrix structure and, on the other hand, occur within the time span defined by these adverbials. Thus, the basic function of (grammatical) inclusive des de-phrases is not to locate the event described in the matrix structure (which they may do, by inference), but to define a temporal frame for the mentioned summation operation. Due to space restrictions, it will not be possible to present here a detailed formal analysis (see Móia 2000 for a proposal within the framework of Discourse Representation Theory, DRT). In section 2,I will present an extended - though succinct - description of the different contexts in which inclusive desde-phrases are sanctioned, which corroborates the hypothesis concerning the distribution of these expressions put forward in section 1. 1. Simple vs. full-scanning inclusive location of events As said, des de-phrases - and their Romance equivalents - have a more restricted distribution than English since-phrases. In particular, their combination with descriptions of single achievements or accomplishments e.g. the counterparts of x die, x get married, x write this book or x built a house - is normally ungrammatical. Sentences below illustrate this fact1: (1)
a. *O Paulo casou desde 1980. t h e P a u l o marry S I M P L E PAST.3rd SG since 1980 "Paulo has got married since 1980." b. *O Paulo escreveu este livro desde 1980. t h e P a u l o write S I M P L E PAST.3rd SG this book since 1980 "Paulo has written this book since 1980."
1 Although I will only consider telic eventualities, it must be noted that (bounded) atelic eventualities behave like telic events in the structures to be described in this paper: (i) O Paulo viveu em Paris duas vezes desde 1980. the P a u l o live S I M P L E PAST.3rd SG in Paris two times since 1980 "Paulo has lived in Paris twice since 1980."
TEMPORAL LOCATION OF EVENTS
139
Some authors (e.g., Heinämäki 1974/1978:86, Kamp & Reyle 1993:631n.64) observe that this kind of sentences may be somewhat odd in English, but is good given an appropriate context. This does not seem to be the case for Romance languages, where, irrespective of the context, sentences like (1) are generally considered ungrammatical. Curiously, there are some contexts where - in contrast to what happens in (1) - the combination of desdephrases with event-descriptions is undisputedly grammatical: (2)
.
b.
Paulo cas ou três vezes desde 1980. the Paulo marrySIMPLE PAST.3rd SG three times since 1980 "Paulo has got married three times since 1980." Paulo escreveu três livros desde 1980. t h e P a u l o write S I M P L E PAST.3rd PL three books since 1980 "Paulo has written three books since 1980."
My contention on these matters (cf. Móia 2000) is that we are dealing with two substantially different modes of temporal location that cross the whole system of adverbial temporal location. Moreover, as shown in examples (1) and (2) above, adverbials of the desde kind are sensitive to this opposition. Examples (1) illustrate what can be termed 'simple inclusive location', expressed by a condition [e t] (where e represents the described event, and t the location time defined by the temporal adverbial). This mode of temporal location - where the basic function of the time adverbial is to provide a frame for locating an event (by stating that it falls anywhere within that frame) - is usually assumed in the literature to be the one associated with event-descriptions (see e.g. Kamp & Reyle 1993:513 - "This seems to be a general property of event-sentences with temporal adverbials such as on Sunday, yesterday, tomorrow morning and many others: they assert that the event falls entirely within the time denoted by the adverb"). Examples (2) illustrate a more complex and interesting location mode, which I will term 'full-scanning inclusive location'. In these sentences, the event described in the matrix structure may be conceived of as a 'complex event' composed of sub-events of the type described (Paulo getting married, or Paulo writing a book, respectively). Formally, in the language of DRT, this event can be represented by a non-atomic discourse referent E, obtained by summation of the relevant sub-events (an analysis that requires the assumption of a Boolean structure for events - see e.g. Bach 1986). Now, the role of time adverbials in these structures appears not to be the same as in the examples (1). In (2), their role is to provide a temporal frame for the above-
140
TELMO MÓIA
mentioned summation operation, assuring that the non-atomic event E is the 'supremum' of the set of relevant sub-events that occur within this frame ('maximality requirement'). Metaphorically speaking, the (location) time associated with these adverbials has to be 'fully-s armed' in order to gather all the relevant sub-events happening within it. In DRT-terms, this location mode can be expressed by the conditions below (where s is the relevant subevent expressed in the matrix structure2, and t the location time, defined by the temporal adverbial):
whence [E
(3) E = ε:
t]
ε
ε
t
1
A fundamental characteristic of the full-scanning inclusive location, which distinguishes it from the simple inclusive one, is that, by providing a frame for abstraction, the temporal adverbial plays a central role in the definition of the eventuality described in the matrix structure. In fact, this eventuality is defined as the set E of (all) subevents ε that, at the same time, correspond to the description in the matrix structure, and happen within t. Thus, if the temporal frame changes, the eventuality E may be different, i.e. have a different composition. A consequence of this fact is that the enlargement of the frame does not necessarily guarantee truth value preservation (this happening, namely, when non-upward-monotone quantification is pertinently involved in the clause) - cf. (4), with an in temporal locator. Obviously, matters are different in structures with simple inclusive location. There, the eventuality described in the matrix structure is defined independently of the location time, which acts as a mere location frame. Hence, if the frame is enlarged, the truth value of the sentence is preserved (obviously due to the transitivity of the inclusion relation) - cf. (5)(6): 2
The abstraction in (3) is recursive, that is, the sub-events Ε in K1 can also be complex events composed by abstraction over other sub-events, as in (i) below if both NPs have a distributive reading: (i) Três actores receberam três Óscares na década de 80. three actors receive,'SIMPLE PAST-3rd PL three Oscars in-the decade of 80 "Three actors received three Oscars in the eighties.'
TEMPORAL LOCATION OF EVENTS (4) (5) (6)
141
Paulo wrote three books in 1985. * Paulo wrote three books in the eighties.3 Paulo got married in 1985. Paulo got married in the eighties. Paulo offered this painting to three friends in 1985. [collective offer] Paulo offered this painting to three friends in the eighties.
In sum, as we can gather from the examples given so far, the contribution of the temporal adverbial is significantly different in the two cases considered. In structures with simple inclusive location, it merely locates the described eventuality; in structures with full-scanning inclusive location, (i) it locates the sub-events and, consequently (by inference), the maximal event E expressed in the matrix structure, and (ii) it contributes, somehow in the manner of a modifier, to defining the maximal event (since the inclusion in the frame associated with the adverbial is a defining property of the elements making up the maximal event). Therefore, it should be stressed that, strictly speaking, what I term 'full-scanning location' appears to be more than a mere mode of temporal location, given the role (ii) above. For simplicity, however, I will use this term (despite the likely inaccuracy), together with the more neutral (and appropriate) term 'full-scanning construction'. Although in this paper I consider only the distribution of the Romance counterparts of since-adverbials, it must be emphasised that sensitivity to the simple vs. full-scanning inclusion appears to be a widespread grammatical phenomenon4. This is only expected, since two distinct functions of temporal adverbials - 'definition of frames for temporal location' vs. 'definition of temporal frames for quantification' - is what is ultimately at stake. It must also be emphasised that the full-scanning construction occurs - provided the right triggers are present - with virtually any temporal locating adverbial, and not only with those that - like des de-phrases - are sensitive to the mentioned distinction. 2. On the distribution of the Romance counterparts of since-adverbials Given that desde-phrases - and their Romance equivalents - permit a clear differentiation of the simple inclusive construction and the full-scanning
3 In the relevant interpretation here, three means exactly three, and the Object NP has a non-specific reading. 4 See e.g. observations on the distribution of since in Heinämäki (1974:86), of until in Declerck (1995:80), or of Portuguese durante [during] in Móia (2000:301).
142
TELMO MÓIA
one, I will take their analysis as the main means of further characterising the latter construction. In particular, I will attempt here to describe the contexts in which it occurs, which is tantamount to identifying the class of 'full-scanning triggers' (i.e. the operators which give rise to the structure in (3)), and some linguistic facts that may block it. For methodological reasons, it is advisable to separate two major types of contexts: those where the des de-phrases occur adverbially within non-subordinate clauses (as in the examples considered so far) and those where these phrases occur within nominal modifers (either adverbially, within clausal nominal modifiers, or adnominally). Let us start with the first. Desde-phrases are grammatical when applied to a matrix structure which contains one of the following seven licensing elements, which I take to be full-scanning triggers: (i) explicit quantifiers over events; (ii) nominal cardinal quantifiers in an NP with distributive (or cumulative) reading; (iii) nominal quantifiers over parts of discrete (or massive) objects; (iv) temporal measure adverbials which apply to atelic descriptions; (v) quantifiers over properties varying on a scale; (vi) exclusion operators; (vii) co-ordination conjunctions associated with distributive readings and exhaustive enumeration - cf. (7)-(13), respectively. (7)
O Paulo casou três vezes desde 1980. the Paulo marrySIMPLE PAST_3rd SG three times since 1980 "Paulo has got married three times since 1980." (8) Très mil pessoas morreram nesta estrada desde 1980. since 1980 three thousand persons dieSIMPLE PAST.3rdPLon-this road "Three thousand persons have died on this road since 1980." (9) Oitenta por cento deste edificio foi destruido desde 1980. eighty per cent of-this building beSIMPLE PAST.3rd SG destroyed since 1980 "Eighty per cent of this building has been destroyed since 1980." (10) Paulo esteve no escritório (durante) mais de cinquenta the Paulo beSIMPLE PAST.3rd SG in-the office (for) more of fifty horas desde segunda-feira. hours since Monday "Paulo has been in the office for more than fifty hours since last Monday." (11) Paulo cresceu muito desde 1980. the Paulo growSIMPLE PAST.3rd SG much since 1980 "Paulo has grown a lot since 1980."
TEMPORAL LOCATION OF EVENTS (12)
Paulo
só
escreveu
este livro desde o ano
143
passado.
t h e P a u l o o n l y write S I M P L E PAST.3rd SG this book since the year past "Paulo has only written this book since last year." (13) Paulo construiu esta casa, esta ponte e esta igreja t h e P a u l o build SIMPLE PAST.3rd SG this house, this bridge and this church desde 1980. since 1980 "Paulo has built this house, this bridge and this church since 1980." Sentence (7) shows the (previously noted) fact that inclusive desde is licensed by the presence of a quantifier over events, such as três vezes [three times]. This quantifier involves explicit reference to a set of events, namely the set of all events of the relevant type (Paulo's weddings) occurring within the location time, and direct specification of its cardinality. Sentence (8) shows that inclusive desde is licensed by the presence of a nominal cardinal quantifier in an NP with a distributive reading. In this example, the licensing NP appears in Subject position, but it might also appear in Object position (as in (2b) above), or in deeper syntactic positions, such as the complement or modifier of a nominal predicate. It must be stressed that this type of sentences - in their distributive reading - involve (at least) as many events as objects relevant for the quantificational process. In fact, we can consider - as is common in the literature - that there is quantification over events also in these cases, albeit an indirect one, via the nominal quantifier. These sentences can therefore be taken to represent a set of events E whose cardinality equals (or is greater than5) that of the relevant set of objects: in (8), for instance, three thousand persons correspond with three thousand dying-events. The crucial fact to note here is that E can again be conceived of as the set of all events of the relevant type occurring within the location time t (which is thus 'fully-scanned'). At this point, it must be stressed that the presence of an NP with a cardinal quantifier is not sufficient to sanction the use of an inclusive desde. For instance, NPs with a group reading apparently do not hold this licensing capacity. Observe the following sentence:
5
If the same object can be involved more than once in the relevant relation, events may outnumber objects. For instance, the sentence Paulo read three books can describe an eventuality comprising six reading-subevents, if each book was read twice. Obviously, matters are different with the sentence Paulo wrote three books, since book-writings are unrepeatable.
144
TELMO MÓIA
(14) Três estudantes compraram three students
um computador nesta loja desde
buySIMPLE PAST_3rd PL a computer
in-this store since a
semana passada. the week past "Three students have bought a computer in this store since last week." Whereas without the desde-phrase this sentence can have both a distributive reading, involving three different buying-events, and a group reading, involving a single corporate buying-event by the three students, with desde it can only have the distributive reading. Formally, I assume, in line with Kamp & Reyle (1993), that NPs with cardinal quantifiers and group readings are not associated with an abstraction operation, contrary to those that have a distributive reading. This amounts to saying that, in such cases, it is existential quantification that is involved. Things being so, it is only natural that the operating monotonicity properties are those of existential quantification, not the (non-)monotonicity of (distributively interpreted) cardinals. That this is the effective inferential pattern for these NPs was already shown above, in (6)6. The incapacity of an NP with a cardinal numeral to license the inclusive desde is also verified when this NP is combined with collective-like expressions such as juntos [together], colectivamente [collectively], ao mesmo tempo [at the same time], or similar ones - cf. (15). An identical blocking effect is obtained, without these collective-like adverbials if, by virtue of the lexical content of the expressions in the sentence (and given our world knowledge), the interpretation involving multiple temporally discontinuous events is excluded - cf. (16).
6
It is interesting to note, in passing, that NPs with cardinal quantifiers and distributive interpretation can exhibit the same behaviour as group NPs concerning the sort of entailment in question and, hence, may not trigger a full-scanning interpretation. This is the case, for instance, if a sentence like (14), without the desde-phrase, is used to describe a (distributional) purchase of computers by three individuals (say, Peter, Susan and Mary) that the speaker has in mind but does not want to name. In these utterance conditions, the relevant NP has a referential (or specific) type of interpretation, and the sentence cannot be assigned a full-scanning reading. Accordingly, the inferential pattern in (6) is valid for these NPs, and they are not licensors of the inclusive desde (cf. similar behaviour of the definite NP estes três estudantes [these three students]).
TEMPORAL LOCATION OF EVENTS
145
(15) *Três pessocLs receberam este premio colectivamente desde 1980. three persons receiveSIMPLE PAST.3rd PL this prize collectively since 1980 "Three persons have won this prize collectively since 1980." (16) *Uma bomba destruiu três edificios desde a semana passada. a bomb destroySIMPLE PAST.3rd SG three buildings since the week past "A bomb has destroyed three buildings since last week." At this point, it is important to stress that, contrary to what these examples might seem to indicate, the sub-events need not temporally distribute over the location time. In fact, they can be simultaneous: (17)
Paulo comprou três fatos desde Janeiro. Por acaso, the Paulo buySIMPLE PAST.3rd SG three suits since January. By chance, comprou-os todos ao mesmo tempo. buySIMPLE PAST.3rd SG them all at-the same time "Paulo has bought three suits since January. Incidentally, he bought them all at the same time."
What seems to cause ungrammaticality is that the simultaneity of the sub-events is explicitly asserted, by way of an adverbial like at the same time, in the same sentence that contains the desde-adverbial, as in (15) (or else, that it is pragmatically coerced, as in (16)). In this case, I hypothesise, the full-scanning is blocked within the sentence with desde, thus causing the observed ungrammaticality. So far, I have only considered distributive and group readings. I will now briefly mention cumulative readings (see Scha 1981), which are also compatible with the inclusive desde: (18) Estes três estudantes compraram dez livros desde o inicio do ano. these three students buySIMPLE PAST.3rd PL ten books since the beginning of-the year "These three students have bought ten books since the beginning of the year." This sentence can describe a situation in which, for instance, the three and - bought a total amount of 5, 2 and 3 mentioned students - , books, respectively, within the relevant period. Obviously, this interpretation requires a full-scanning of the location time (in order to count all the book-buying events involving each of the three mentioned students). The
146
TELMO MÓIA
following entailment impossibility furthermore shows that these 'cumulative structures' behave like those with distributive NPs (and unlike those with group NPs), with respect to the location time impact on defining the main clause eventuality: (19) These three students bought (exactly) ten books on May 15. * These three students bought (exactly) ten books in May. The three contexts illustrated in (9)-(ll) contain different types of quantifying elements, which, unlike those observed up to now, express some form of measurement rather than counting, the quantified entities being: parts of discrete 'ordinary' objects, the duration of (atelic) eventualities, and properties varying on a scale, respectively. Still, as in the previous cases, the sentences with these elements involve a full-scanning of the location time (which sets a temporal frame for quantification in like manner). In fact, the sentences in question can be taken as representations of the set of all subevents of a given type (involving parts of the relevant entities) that occur within the location time t. Furthermore, the restrictions on the licensing capacities of the quantifiers and the blocking effects are parallel to those observed before: (20) * Oitenta por cento deste edificio foi destruido de urna Eighty per cent of-this building be SIMPLE PAST-3rd SG destroyed o f one só vez desde 1980. only time since 1980 "Eighty per cent of this building has been destroyed all at once since 1980." It is quite interesting to note that temporal measure adverbials that quantify the duration of telic eventualities, headed by the preposition em [in] in Portuguese, do not license the use of the inclusive desde (contrary to those that quantify the duration of atelic eventualities): (21) * 0 Paulo leu este livro em duas horas desde ontem. the Paulo read SIMPLE PAST_3rd SG this book in two hours since yesterday "Paulo has read this book in two hours since yesterday." Now, this difference in the licensing capacity of durante [for] and em [in] parallels the following one, concerning entailments of the type presented
TEMPORAL LOCATION OF EVENTS
147
above in (4)-(6), which assess whether the location frame contributes to define the main clause eventuality: (22) Paulo read this book in two hours on May 15. —» Paulo read this book in two hours in May. (23) Paulo worked for two hours on May 15. *—» Paulo worked for two hours in May. What this contrast seems to indicate is that durante / for measure phrases are full-scanning triggers, whereas em / in measure phrases are not. In other words, the sentences where the latter occur with a temporal locating adverbial involve simple inclusive location (obviously, if no other full-scanning trigger occurs in it). Let us now consider sentence (12) - repeated below as (24) - which shows that the inclusive desde can occur in structures with só, the Portuguese counterpart of only. Note that the exclusion operator may appear in combination with other licensing expressions (e.g. NPs with cardinal quantifiers and distributive reading), but this need not be so, which is the relevant case to consider here. (24)
Paulo só escreveu este livro desde o ano passado. "Paulo has only written this book since last year."
This sentence - ungrammatical without the exclusion operator - has different readings, depending on which constituent is focussed. Let us consider three possibilities: (i) focus on this: "the set of books that Paulo has written since last year is formed only by this book" (he may have written other things e.g. papers - within this period); (ii) focus on this book: "the set of things that Paulo has written since last year is formed only by this book"; (iii) focus on written this book: "the set of (relevant) things that Paulo has done since last year is formed only by the event of writing this book". The paraphrases just given evince the parallelism between these constructions and those with (singular) cardinal quantifiers7, showing that a full-scanning of the location time also operates here. I will not develop the analysis of these structures, since the study of exclusion operators is a topic of its own, but I assume that a representation along the lines of (3) can be used for these cases as well, and is
7
Constructions with plural expressions - e.g. estes n livros [these n books'] instead of este livro [this book] - are of course also possible.
148
TELMO MÓIA
furthermore in line with the analyses proposed in the literature for exclusion operators (cf. in particular Room's 1985 semantics for focus). The seventh context I considered involves nominal, verbal or sentential co-ordination, again associated with a distributive reading. It is illustrated for nominal co-ordination - in (13), repeated below as (25). (25)
Paulo construiu esta casa, esta ponte e esta igreja desde 1980. "Paulo has built this house, this bridge and this church since 1980."
In my opinion, this Portuguese sentence is only indisputably grammatical if it represents an 'exhaustive enumeration' of the relevant events that happened within the interval defined by the desde-phrase (exhaustiveness emerging possibly as an implicature). For instance, in a scenario where Paulo is an architect, the above sentence seems to be (indisputably) felicitous only if the named three works are all Paulo accomplished within the mentioned period; should he have built a baseball stadium, for instance, during that time, the sentence would be an inaccurate - insufficient - description of the facts. Accordingly, (25) is paraphrasable as "the set of (all) things that Paulo has built since 1980 is formed by this house, this bridge and this church" or, attesting the similarity with contexts illustrated in (8) and (12) above, "Paulo has (only) built three things since 1980: this house, this bridge and this church". I will now briefly consider sentences where the inclusive desde-pbrasos occur within nominal modifiers. In these sentences, the matrix structure need not contain any of the seven licensing elements described above. In fact, they may appear: (i) in plural definite nominals, both in adverbial position within relative clauses and in strict adnominal position - cf. (26)-(27), respectively; or (ii) in singular or plural nominals which contain ordinals, superlatives or a counterpart of the only - cf. (28) and (31)-(32): (26) Muitos dos livros que Paulo escreveu desde 1980 many of-the books that the Paulo writeSIMPLE PAST.3rd SG since 1980 foram premiados. beSIMPLE PAST.3rdPLawarded-a-prize "Many of the books that Paulo has written since 1980 were awarded a prize."
TEMPORAL LOCATION OF EVENTS
149
(27) Todos os golpes de estado na América Latina desde 1980 all the coups d'état in-the America Latin since 1980 foram perpetrados pelos militares. beSIMPLE PAST.3rd PL perpetrated by-the militaryN0N,C0LLECTIVE_PL "All the coups d'état in Latin America since 1980 were carried out by the military." (28) Este é terceiro livro que Paulo escreveu desde 1980. this beSIMPLE PRESENT_3rd SG the third book that the Paulo writeSIMPLE PAST.3rd SG since 1980 "This is the third book that Paulo has written since 1980." In (26), the use of the inclusive desde is grammatical - in adverbial position despite the absence of a licensing operator like those described above within the relative clause (though it may of course be present, as in the counterpart of many of the books that Paulo has read three times since 1980 were awarded a prize). In (27), there is also no licensing operator of any type mentioned within the nominal constituent containing the desde-phrase. The grammaticality of (26)-(27) in the absence of such full-scanning triggers is accounted for by the fact - so I hypothesise - that the desde-adverbial itself is processed, in these contexts, within a sub-DRS created by abstraction. In fact, the bracketed structures in these sentences are of the type: (29) [N. N[+PLURAL] MOD] (where the locating adverbial occurs inside MOD - a relative clause - in (26), and is MOD, in (27)). Plural nominal constituents often represent the sum of all entities corresponding to the descriptive content of N that have the property expressed by MOD, e.g. all the books that Paulo wrote between 1980 and the 'temporal perspective point', in (26), and all the coups d'état occurred in Latin America between 1980 and the 'temporal perspective point', in (27). Hence, the distribution of inclusive desde in these constructions is accounted for by the same principle as those before, to wit: inclusive desde is only sanctioned in contexts where the interval it defines is fully scanned. Conversely, singular nominal constituents do not, as a rule, license the use of inclusive desde (probably due to the fact that they do not normally require an abstraction of the type described above):
150
TELMO MÓIA
(30) a.
*Conheço um livro que Paulo escreveu desde 1980. [null Subject] knowSIMPLE PRESENT.lst SG a book that the Paulo W r i t e SIMPLE PRESENT-3rd SG S i n c e 1 9 8 0
vs.
b.
"I know a book that Paulo has written since 1980." Conheço {todos) os livros que Paulo escreveu desde 1980. [null Subject] knowSIMPLE PRESENT.lst SG (all) the books that the Paulo writeSIMPLE PRESENT.3rd SG since 1980 "I know (all) the books that Paulo has written since 1980."
Sentence (28) illustrates a context with a sanctioned inclusive desdeadverbial where the relevant N'-node is singular, and where none of the triggers considered in (7)-(13) above is present. This structure is, however, not extraordinary, since ordinals are intuitively full-scanning triggers: in order to determine whether the mentioned book was the third to be written within the mentioned time span, this span has to be fully scanned. Other singular constructions with comparable properties are those containing superlatives, and o único [the only]: (31) Este é o livro mais interessante que o Paulo escreveu desde 1980. this beSIMPLE PRESENT.3rd sG the book more interesting that the Paulo writeSIMPLE PAST.3rd SG since 1980 "This is the most interesting book that Paulo has written since 1980." (32) Este é o único livro que o Paulo escreveu desde 1980. this beSIMPLE PRESENT.3rd SG the only book that the Paulo writeSIMPLE PAST.3rd SG since 1980 "This is the only book that Paulo has written since 1980." 3.
Conclusions The facts considered in this paper illustrate an intriguing interaction between the system of adverbial temporal location and that of quantification, which - as far as I know - had not been explored in the semantic literature. In fact, the full-scanning construction involves a special use of temporal 'locating' adverbials in which the locating function is, as it were, subsidiary. Here the import of these adverbials lies essentially in the definition of a temporal frame for some quantificational operation - e.g. counting eventualities, determining the total amount of time consumed in a given activity, exhaustively enumerating relevant events, comparing properties of objects, or simply picking up maximal sums of entities to be involved in
TEMPORAL LOCATION OF EVENTS
151
verbal predication (cf. (26)-(27) for the latter case). Normally, these quantificational operations are not temporally unbounded, but rather circumscribed in time, and relative to the span taken into account. Now, the temporal circumscription at stake is expressed par excellence via a time adverbial. Formally, in DRT-terms, the time adverbial sets the frame for quantification, in these constructions, by characterising a time discourse referent t which occurs in a sub-DRS under the scope of an abstraction operator. Accordingly, in a combination main clause / locating adverbial where full-scanning takes place, the temporal adverbial contributes to defining the main clause non-atomic eventuality E, given that the inclusion in the frame set by the adverbial is a constitutive property of the sub-events making up E. For most temporal adverbials, the difference between the simple temporal location and the full-scanning one is not easily perceptible. Interestingly, though, there are adverbials which occur in one of the contexts but not in the other, allowing a clearer demarcation of the full-scanning construction. The case I explored in more detail is that of Portuguese desdeadverbials, which, like their counterparts in other Romance languages, are only compatible with the full-scanning location, in combination with event descriptions.
REFERENCES Bach, Emmon. 1986. "The Algebra of Events". Linguistics and Philosophy 9. 5-16. Declerck, Renaat. 1995. "The problem of not...until". Linguistics 33.51-98. Heinämäki, Orvokki. 1974. Semantics of English Temporal Connectives. Ph.D. Dissertation, University of Texas, Austin. Reproduced by Indiana University Linguistics Club, 1978. Kamp, Hans & Uwe Reyle. 1993. From Discourse to Logic. Introduction to Modeltheoretic Semantics of Natural Language, Formal Logic and Discourse Representation Theory. Dordrecht: Kluwer. Móia, Telmo. 2000. Identifying and Computing Temporal Locating Adverbials with a Particular Focus on Portuguese and English, Ph.D. Dissertation, Universidade de Lisboa. Rooth, Mats. 1985. Association with Focus. Ph.D. Dissertation, University of Massachusetts. Reproduced by GLSA Publications, University of Massachusetts, Amherst.
152
TELMO MÓIA
Scha, Remko. 1981. "Distributive, Collective and Cumulative Quantification". Formal Methods in the Study of Language ed. by Jeroen A. G. Groenendijk, T. Jansenn and Martin Stokhof, 483-512. Amsterdam: Mathematisch Centrum. Vlach, Frank. 1993. "Temporal Adverbials, Tenses and the Perfect". Linguistics and Philosophy 16.231 -283.
PRO, MOVEMENT AND BINDING IN PORTUGUESE* ACRISIO PIRES University of Maryland at College Park
0.
Introduction Portuguese infinitives are especially relevant for a discussion about control because, unlike in most other languages, they may display overt morphological marking for person and number agreement. Section 1 provides evidence that non-inflected infinitives display interpretive properties of obligatory control ( ) and inflected infinitives allow non-obligatory control (NOC) properties. In section 2 I argue in favor of an A-movement analysis of different kinds structures in Portuguese: control by the matrix subject, control by the of object and adjunct control. The first advantage of this analysis is that it accounts for the morphological split in the inflectional morphology of infinitives and its syntactic effect in different dialects of Portuguese, offering a principled explanation for the control interpretive contrasts discussed in section 1. In section 3 I address the case of non-overt subjects of inflected infinitives, arguing that what licenses their (NOC) subjects is their inflectional morphology. Section 4 provides independent evidence from binding supporting the movement approach defended here. Section 5 addresses dialects of BP that have lost the inflected infinitives, and explains their behavior given the analysis proposed for the dialects that preserve these constructions. In Section 6 I briefly address cases of arbitrary control PRO, I am grateful to Norbert Hornstein, David Lightfoot and Juan Uriagereka for useful suggestions and criticism on this research. Thanks to the participants at the Univ.of Maryland Student Conference in Linguistics (Nov 99), LSRL XXX, 10th CGG (Alcalá, Madrid, Apr 00), two anonymous reviewers and other people who provided suggestions or judgments on the data, specially: Joäo Costa, Enilde Faulstich, Maria Joäo Freitas, Sónia Frota, Ana Gouvea, Max Guimaräes, Allan Munn, Telmo Móia, Rozana Naves, Marcia Oliveira, Cilene Rodrigues, Itziar San Martin, Cristina Schmitt, Adriana Viana and, finally, Lucia Lobato, who helped me to contact informants. Any inaccuracies remain my responsibility. This work was partly supported by a fellowship from Brazil's Dept. of Education (MEC/CAPES Brasilia).
154
ACRISIO PIRES
distinguishing them from the PRO cases analyzed here as the result of A-movement. I consider data from three dialects of Portuguese: European Portuguese (EP), Standard Brazilian Portuguese (StdBP, which is represented by the language used in the media) and Colloquial Brazilian Portuguese (ColBP). 1. Control in dialects with inflected infinitives Infinitives in Portuguese come in two varieties: non-inflected or inflected for person/number. Standard Brazilian Portuguese in general displays the person/number paradigm for inflected infinitives shown in (1), where the plural forms display overt inflectional morphology for person/number: (1) (StdBP)1 SG 1 (eu) fala-r 2 (você) fala-r 3 (ele/ela) fala-r PL 1 (nós) fala-r-mos 2 (vocês) fala-r-em 3 (eles/elas) fala-r-em
"(I) speak-INF-o" "( ou) speak-INF-ø" "(he/she) speak-INF- 0" " ( w e ) speak-INF-lPL" "(you-PL) speak-INF-3PL" "(they) speak-INF-3PL"
EP differs from the paradigm in (1) only by the fact that it has specific inflected forms for 2nd person singular and 2nd person plural, given the possible use of different pronouns with those forms. Since only 1st and 3 rd person plural verb forms display the same overt morphology for person/number in both EP and StdBP, I will restrict my examples of inflected infinitives to those forms. How do Portuguese Infinitives behave with respect to control? Only noninflected infinitives consistently show properties of obligatory control ( ). This can be seen when one applies the diagnostics for control discussed in Hornstein (1999) and based on Lebeaux (1985), as follows. 1.1 Need for a local c-commanding antecedent First, consider the need for a local c-commanding antecedent in examples (2a) and (3a). PRO in the embedded infinitive clauses in (2a from EP) and (3a from StdBP) must have a local antecedent in the matrix clause. That antecedent must also c-command PRO. So, both in (2a) and (3a) the entire
1 INF stands for the infinitive morpheme, followed in the plural forms by the overt inflection for person + number (1P/3P).
PRO, MOVEMENT AND BINDING IN PORTUGUESE
155
bracketed DP in the matrix clause is the only possible antecedent for PRO in the embedded clause. (2)
a.
b.
(3)
a.
b.
[Os pais do Paulojk lamen tam PRO*/ chegar [the parents of-the Pauloj]k regret PRO*j/k arrive-INF tarde. (EP) late. "[Pauloj's parents]k regret PRO*j/k to arrive late." [Os nossoSj pais ] lamentam proj chegarmos tarde. [the ourj parents] regret proj arrive-INF-lPL late. "Our parents regret our arriving late." [NossoSj amigos]k detestam PRO*j/k perder as coisas [ourj friends]k hate PRO*j/ k lose-INF the things deles. (StdBP) of-they. "Our friends hate losing their belongings." [NossoSj amigos] detestam proj perdermos as cois as [ourj friends] hate proj lose-INF-lPL the things deles. of-they. "Our friends hate when we lose their belongings."
A local c-commanding antecedent is not needed for the subject of the embedded infinitives in (2b) and (3b), which are inflected. With inflected infinitives pro can also be disjoint in reference from any DP in the sentence, at least with 1PL inflection. This and the other contrasts below show that inflected infinitives display NOC properties, whereas properties are restricted to non-inflected infinitives. 1.2 Sloppy reading under ellipsis The ellipsis material in (4a) and (5a) (indicated between parenthesis) only allows a sloppy reading. For instance, when the ellipsis material regrets to have lost is interpreted in (4a) it can only mean that Silvia herself regrets her own losing. This contrasts directly with examples (4b) and (5b). Take (4b), where the second conjunct must be interpreted as Silvia regrets our losing, corresponding to a strict interpretation of the ellipsis site.
156
ACRISIO PIRES
(4)
.
b.
(5)
a.
b.
Pauloj lamenta PROj/ *k ter perdido e a Silvia the Pauloj regrets PROj/ *k have-INF lost and the Silvia também. (= lamenta ter perdido) (EP) too. (= Silvia regrets to have lost) "Pauloj regrets PROj/ *k to have-INF lost and Silvia does too. (= Silvia regrets to have lost) " Pauloj lamenta prok termos perdido e a thePauloj regrets prok have-INF-lPL lost and the Silvia também. (= lamenta nós termos perdido) (EP) Silvia too (=Silvia regrets our losing) "Paulo regrets our losing and Silvia does too. (=Silvia regrets our losing)." O Pedroj aceita PROj/ *k votar todas as propostas the Pedroj accepts PROj/ *k vote-lNF all the proposals hoje e a Ana também. (=Ana aceita votar todas as propostas hoje) (StdBP). today and the Ana too. (=Ana accepts to vote all the proposals today) "Pedro accepts to vote all the proposals today and Ana does too. (=Ana accepts to vote all the proposals today)." O Pedroj aceita prok votarmos todas as propostas the Pedroj accepts prok vote-INF-lPL all the proposals hoje e a Ana também. (=Ana aceita que nós votemos todas as propostas hoje). (StdBP) today and the Ana too. (=Ana accepts that we vote all the proposals today) "Pedro accepts that we vote all the proposals today and Ana does too. (=Ana accepts that we vote all the proposals today)."
Finally, the next section presents a third test distinguishing the properties of non-inflected infinitives from the NOC properties: 1.3 Impossibility of split antecedents Notice how the two types of infinitive behave with respect to the possibility of split antecedents for the embedded clause null subject. PRO in the non-inflected infinitive in (6a) rejects an interpretation in which both I and Maria form a set that behaves as the antecedent for PRO. (6b) is clearly distinct in that respect. Here pro is -referential with a set of elements that can include both I and Maria.
PRO, MOVEMENT AND BINDING IN PORTUGUESE (6)
a.
b.
157
Euj convencí a Marian PROk/*j+k a viajar com Paulo. (EP & STBP/some informants) Ij convinced the Mariak PROk/*J+k to travel-INF with the Paulo. "Ij convinced Mariak PROk/*j+k to travel with Paulo", EUJ convencí a Mariak roj+ a viajarmos como Paulo. Ij convinced the Mariak proj+k,to travel-INF-1PL with the Paulo. "Ij convinced Mariak that wej+k (should) travel with Paulo."
None of the properties discussed above holds for the subject of an embedded inflected infinitive. However, they must apply for non-inflected infinitives. Hornstein (1999) used these different properties as arguments for an analysis of control in English as the result of DP movement. Take for instance the impossibility of split antecedents. If PRO is the result of DP movement from the embedded clause to an appropriate position in the matrix clause, split antecedents are not possible, because a DP cannot move to two different positions at the same time. In the next section, I present the details of an analysis for control in Portuguese along the lines proposed by Hornstein (1999). 2. Null subjects of uninflected infinitives as the result of DP-movement In the analysis below I follow proposals by Boskovic (1994), Boskovic and Takahashi (1998), Hornstein (1999) and Lasnik (1995) and assume that multiple 9-roles can be discharged on DPs, as the result of movement, contra Chomsky (1995, 1998, 1999). If one denies the existence of a D-structure level (Chomsky 1993; see also Uriagereka 1998 for details and references), there seems to be no strong argument against the possibility of having multiple thematic roles assigned to the same DP in the course of the derivation. I assume here that 0-roles are features. That is consistent with the idea that movement is motivated by feature checking. However, that is not crucial for this analysis, and 0-role checking could also take place configurationally (cf. Boeckx 2000 for such a proposal). I follow the approach to feature checking proposed in Chomsky (1998, 1999). A T° carries a set of [-interpretable] -features that needs to be checked by an agreeing DP. Only a T° that carries a complete set of -features (i.e. which is -complete) can check the Case of an atracted DP. Finally, any T° carries a [-interpretable] EPP feature that can be checked by the categorial
158
ACRISIO PIRES
feature of an atracted DP. Therefore, a DP is allowed to check the EPP and the -features (that is, at least person and number) of more than one T°. That is consistent with Chomsky's (1999) view that an embedded (non-inflected) infinitive has a defective head T°, that is, it doesn't have a complete -set, although it carries an [-interpretable] EPP-feature. Therefore, this defective T is unable to determine Case-agreement. Since it has an EPP feature, it can attract a DP to its Spec, but it cannot delete the [-interpretable] Case feature of the attracted DP. 2.1 Subject control Consider a case such as (7) where only a non-inflected infinitive is licensed in Portuguese: (7)
nósj/*K [VP¡nós sair]]]].2 [Tpi wej/*k [Vpiwe leave]]]].
[TP2NÓSJ [VP2 nós conseguimos [TPI
[TP2Wej [vp2 we managed "We managed to leave."
The subject of the matrix clause in (7) is first merged in the embedded clause, where it checks the theta-feature of VPI and the EPP-feature in the Spec of TP1, but cannot have its Case checked, because [Spec, TP] of a non-inflected infinitive clause is not a Case-checking position. Therefore, it must be raised to the matrix clause where it checks the 8-feature of VP2, the EPP-feature of TP2, and its own Case feature, freezing in [Spec, TP2]. Once all the features have been checked the lower copies are deleted according to the copy theory of movement (Chomsky 1995, Nunes 1995) and the derivation converges. 2.2 Object control In this section I derive a case of object control assuming the movement analysis used above. In example (8) Maria checks the theta-role of the embedded VP and then moves to [Spec, TP] of the embedded clause, where it checks an EPP feature. According to Hornstein (1999), this kind of movement violates Procrastinate', since I, which is still in the numeration, could have been inserted in [Spec, TP] of the embedded clause. Maria then moves to the VP of the matrix clause, which is another violation of 'Procrastinate', given 2 1 represent Portuguese infinitives as TPs to simplify the notation. However, they may in fact be CPs, although that is not crucial here. I do not address here the optional fronting of the verb and facts about wh-movement discussed in Raposo (1987), who argues for a distinction between TP and CP for inflected infinitives in EP. Galves (1992) proposes that they are always CPs.
PRO, MOVEMENT AND BINDING IN PORTUGUESE
159
that I is still in the numeration. Both violations are possible because otherwise the derivation would not converge. (8)
Eu convencí a Maria [a Maria a viajar hoje]. I convinced the Maria [the Maria to travel-INF today]. "I convinced Maria to travel today."
The existence of 'Procrastinate' as an independent principle has been recently challenged by Chomsky (1999). Furthermore, the effect holds only if one assumes that an item in the numeration can merge in a non-thematic position. If first merge is restricted to thematic positions (a weaker version of 'pure merge'), the problem does not arise. Let us assume that the restriction on 'first merge' does not hold, in order to consider a couple of consequences. If it is correct that 'merge' is more economical than 'move', the effect described above is still present. How to eliminate this problem? In one alternative, following Hornstein (1999), Maria is inserted in the derivation with an accusative Case feature, and I with a nominative Case feature. This prevents I from merging into the accusative Case position of the matrix clause, forcing Maria to move to that position. Nevertheless, if we assume that nominative and accusative Case features are specified by -feature agreement in the course of the derivation (Chomsky 1998, 1999), another explanation is necessary for the violation of 'merge over move' in this Case. We appeal to the Minimal Link Condition to account for that. If I is merged in [Spec, TP] of the embedded clause (satisfying 'merge over move'), instead of moving Maria from the embedded VP, /would block the attraction of Maria both to the matrix object and subject positions. Therefore, the Case of Maria wouldn't be checked and the derivation would crash. 2.3 Adjunct control I present below a derivation for a case of adjunct control. First, notice that an object can bind, and thus c-command, a pronoun within an adjunct, as shown in (9): (9)
John read every bookj without reviewing itj.
In a sentence like (10) below, the adjunct PP [before Pedro leaving] is built first in the derivation. At the point at which TP1 is built, the main clause starts out as a separate derivation, on a parallel space. Maria merges with called in VP2. Pedro, which is in the parallel derivation of the non-finite
160
ACRISIO PIRES
clause can move by sidewards movement (following Nunes 1995) into the matrix external θ-role position. That is possible only because at that point in the derivation the adjunct has not merged with the matrix clause yet. After moving into the matrix clause, Pedro checks the external theta-role of called and moves to [Spec, TP2] where it checks the EPP feature and its own Case feature: (10) [TP2 Pedro [VP2 chamou a Maria [pp antes de [TPI e Pedro [TP2 the Pedro [vp2 called the Maria [PP before of [TP1 the Pedro [VPI the Pedro sair]]]]]. [VPI the Pedro leave-INF]]]]]. "Pedro called Maria before leaving." Three points need to be explained with respect to the derivation above. First, if Maria had been inserted in [Spec, TP] of the embedded clause instead of moving Pedro to that position, Pedro would have never checked its Case feature, by virtue of the Minimal Link Condition (MLC). Second, if Pedro moved to the matrix object position instead of merging Maria, there would be a violation of 'merge over move'. Finally, moving Pedro to the external theta-role position of the matrix over Maria does not involve a violation of 'merge over move', since both DPs are already in the derivation. Also, the MLC does not play any role in this case. Neither Pedro nor Maria c-command each other, and the target does not c-command them both. Since the MLC and locality can only be measured in terms of c-command, both DPs are equidistant from the target. However, if Maria moved to the external thetarole position, Pedro would not check its Case. 3. Non-overt subjects of inflected infinitives Inflected infinitives in Portuguese, contrary to uninflected ones, carry a complete -set. That way, T° can delete the Case feature of the attracted DP, forcing it to freeze in the embedded clause subject position. Consider example (11): (11) [TP2 A Mariaj [Vp2 a—Maria chegou [PP antes de [Tp1 prok [Vp1 prok [TP2 the Mariaj [Vp2 the Maria arrived [PP before of [TP1\ r k [VP1 r k sairmos]]]]]. leave-INF-1PL]]]]]. "Maria arrived before we left."
PRO, MOVEMENT AND BINDING IN PORTUGUESE
161
I keep the standard analysis for the null subject of inflected infinitives in Portuguese, by arguing that it is a pro (Raposo 1987 and references therein). Under this view, the morphological distinction between inflected and non-inflected infinitives correlates directly with a movement analysis of the control cases represented by non-inflected infinitives. T° of an inflected infinitive clause carries a full set of -features (as indicated by the overt person/number morphology), which can check the Case feature of a pro or overt DP occurring in the subject position of the infinitival clause, blocking any further movement (cf. Chomsky 1998, 1999). In (11) pro merges in VPl where it checks one theta-feature. It is then attracted by T° in TP1. T°, which is -complete in this case, has its own EPP-feature deleted, and also deletes the Case feature of pro. That way, pro is frozen in the embedded clause, and the derivation converges. Now consider (12), which is ungrammatical in EP and StdBP. Since T° of a non-inflected infinitive cannot check the Case feature of a lexical subject, the derivation crashes: (12)
Maridj [VP2 chegou [PP antes de [Tp1nós [vpsair]]]]]. [TP2Mariaj [Vp2 arrived [pP before of [Tp1 we [Vp leave-INF]]]]]. "*Maria arrived before we to leave." *[TP2
Given that T° of non-inflected infinitive does not have a complete set of -features, Case cannot be checked in the embedded [Spec, TPl], forcing the subject DP to move to the matrix clause to satisfy its Case requirement. However, if the NP we moved to the matrix clause, the NP Maria would remain in the numeration, that is, it would not enter the derivation although it had already been selected from the lexicon, and the derivation would be cancelled. Notice that this analysis also holds for a case like (12), but with a pro in the embedded subject position.3 4. Evidence from binding I present below evidence from binding in Portuguese that supports an analysis of control as the result of movement. Consider example (13). The non-inflected infinitive constructions where PRO is required are also cases where the matrix clause is the appropriate domain for binding an anaphor that occurs in the embedded infinitive clause:
3
San Martin (1999) analyzes cases of control in Basque that display a contrast similar to the one in Portuguese, involving a split in terms of Case marking instead of inflection.
162
ACRISIO PIRES (13) Elesj aceitaram sej barbear(*em) sem espelho. (BP) Theyj accepted self] shave-INF-(*3PL) without mirror. "Theyj managed to shave (themselvesj) without a mirror."
Inflected non-finite clauses as in (14), on the other hand, behave as independent domains for anaphor binding, forcing an anaphor to be bound by pro or a lexical DP in the subject position of the embedded clause: (14)
diretorj começou a cerimônia antes de nós/prok nosk the director] started the celebration before of we/prok ourselvesk apresentar*(mos). (StdBP). introduce-INF- * ( 1 PL). "The directorj started the celebration before we introduced ourselves."
These facts support the argument that PRO is the result of movement, if we assume the standard view (based on arguments by Lebeaux 1985, and Chomsky 1993) that anaphors raise to the domain of their antecedent to satisfy their binding properties at LF. Cases of pronoun binding in Portuguese can also be used as evidence for a movement analysis of PRO, but only to a certain extent. For most informants, in ColBP an overt pronoun in an infinitive clause cannot be co-referential with an antecedent in the matrix clause, as shown in example (15): (15) A Reginaj vai esperar ela*j vender o carro. (ColBP) The Reginaj will wait her*j, sell the car. "Reginaj will wait for her*j,k to sell the car." However, the embedded clause in (15) should actually be analyzed together with the inflected infinitive examples (16) and (17), where there is the same requirement for disjunction in reference between the embedded and matrix subjects: (16) [O Pedro e Paulo]j sairam depots deles y k [The Pedro and the Paulo]j left after of-they*j/k 1er*(em) o jornal. (EP and BP). read-INF-*(3PL) the paper. "[Pedro e Paulo]j left after they*j/k read the paper."
PRO, MOVEMENT AND BINDING IN PORTUGUESE
163
(17) [O Pedro e Paulo]j vão esperar eles*j/k vender(em) [The Pedro and the Paulo]j will wait they*j/k sell-lNF-(3PL) the carro. (ColBP/some informants) car. "[Pedro e Paulo]j will wait for them*j/k to sell the car." In fact, not only the non-inflected infinitive in (15), but also the inflected infinitive domain in (16) and (17) prevent co-reference between the embedded subject and the matrix one. But this is so not because the matrix clause should function as the binding domain for the embedded clause in both types of example. If there is person/number agreement in the embedded clause, it should actually function as an independent domain for binding, as already shown in example (14) for anaphor binding. Rather, in cases such as (16) and (17), other factors are at play, causing an obviation effect: obligatory disjunction in the interpretation of the embedded pronoun with respect to the matrix clause. In that respect, the pronouns in the inflected infinitives in (16) and (17) cannot be bound within the embedded clause (that is, their binding domain) by virtue of the binding theory. Furthermore, due to factors involving obviation effects, these pronouns cannot be bound in the matrix clause either, showing an effect that is very much like what happens to subjunctives in several Romance languages. Kempchinsky (1986) proposed an account for such effects in subjunctives, and recently Hornstein and San Martin (2000) have tried to account for similar facts using an economy approach. 5. BP dialects with loss of inflected infinitives This section summarizes evidence from language change that supports a tight relation between overt inflection and the licensing of NOC properties in Portuguese (see Pires, forthcoming, for details). Certain dialects of Colloquial Brazilian Portuguese (ColBP) have lost the overt inflectional morphology for person/number in inflected infinitives, given two facts: i.
ii.
The form a gente "the folks" has replaced nós "we" in subject position, and its agreement morphology corresponds to 3SG, although it refers to 1PL in the discourse, as shown in the gloss in (18). The 3PL marking for person/number is used less and less, and has disappeared entirely from some dialects, together with the 1PL marking from dialects that still license the form nós "we".
164
ACRISIO PIRES
SG PL
1 (eu) falar 2/3 (você/ele/ela) falar 1 (nos) falarmes 1 (a gente) falar 2/3 (vocês/eles/elas) falarem
"(I) speak-INF-0" "(You/he/she) speak-INF-0' "(We) speak-INF-4Pfc" "(We) speak-INF-0" "(You/they) speak-INF'
This morphological simplication has a clear syntactic effect: properties of NOC are not licensed with infinitives in ColBP in the cases where the overt person/number morphology has disappeared. Consider example (19) from ColBP, where the 1st person plural verb form for the inflected infinitive is no longer used. The 1 st person plural we and its corresponding verb form have generally been replaced by the form a gente (literally "the folks", but interpreted as "we"). (19) Euj convencí a Mariak [PROk/*j+k a viajar(*mos) hojej. (ColBP) Ij convinced the Mariak [PROk/*j+k to travel-INF(*lPL) today]. "Ij convinced Mariak PROk/*j+k to travel today." Although the morphology of the embedded infinitive is compatible with the current 1PL pronoun form a gente "the folks", the subject of the embedded infinitive in (19) cannot be interpreted that way. The control tests discussed in section 1 hold for such cases, indicating that they display properties. That shows that the lack of overt agreement morphology forces the embedded subject to behave as a trace of DP movement in ColBP, supporting the analysis proposed in sections 2 and 3, according to which the presence of overt inflectional morphology (and thus a complete set of -features) is necessary to license a non-overt subject with NOC properties.4 6. Arbitrary PRO Non-obligatory control (NOC) properties occur not only with inflected infinitives, but also with non-inflected infinitives that allow an arbitrary PRO interpretation (ArbPRO, so called in standard Control Theory. Cf. Martin 1996 and references therein). There is no need for an antecedent in ArbPRO PRO do constructions such as (20). Other diagnostics discussed here for not hold either, as a consequence of the lack of an antecedent.
4
Modesto (1999) and Rodrigues (forthcoming) analyze the loss of referential pro in finite domains in Brazilian Portuguese.
PRO, MOVEMENT AND BINDING IN PORTUGUESE
165
(20) E impossível PROk bater esse record. (EP and BP)5 is impossible PROk beat-INF this record. "It is impossible to beat this record." This might suggest that subjects of non-inflected infinitives with ArbPRO interpretation should be conflated with arbitrary plural subjects as in (21), which also display NOC properties. Another similarity between the two cases is that they both allow either a singular or a plural interpretation for the null arbitrary subject. (21) E impossivel prok haterem esse record. (EP and BP) is impossible prok beat-INF-3PL this record. "It is impossible to beat this record." However, there are several distinctions between (20) and (21). For instance, whereas in (20) the speaker can be included in the reference set for the null arbitrary subject, in (21) the speaker is excluded from that reference set. This and other distinctions between both types of arbitrary subjects have been discussed by Jaeggli (1986).6 What is relevant for the discussion here is that whatever properties distinguish these arbitrary subjects from the non-arbitrary subject pro of inflected infinitives discussed in section 3, they all share the non-obligatory control interpretation properties discussed in section 1. This sets all of them apart from obligatory control PRO, which has been analyzed here as the result of DP movement. 7.
Conclusion I analyzed obligatory control as DP movement in three Portuguese dialects, following Hornstein (1999). First, the proposal presented above has the advantage of providing visible morphosyntactic evidence supporting this kind of analysis. Obligatory control PRO is eliminated as an independent element in the grammar. This provides motivation for dispensing with the PRO Theorem and the Control Module. The minimalist reasoning is that all things being equal, the fewer the modules the better. Furthermore, the current analysis captures in a straightforward way the contrast between obligatory and non-obligatory control, and its connection to the split in the inflectional morphology of infinitives in a language like Portuguese. 5
BP applies to both dialects of Brazilian Portuguese discussed here. An anonymous reviewer pointed out the analysis by Jaeggli (1986). Cristina Schmitt (p.c.) also called my attention to a possible distinction between both types of NOC subjects.
6
166
ACRISIO PIRES
REFERENCES Boeckx, Cedric. 2000. "Notes on missing object constructions". Ms. University of Connecticut, Storrs. Boskovic, Zeljko 1994. "D-Structure, Theta Criterion, and movement into theta positions". Linguistic Analysis 24.247-286. Boskovic, Zeljko, and Daiko Takahashi. 1998. "Scrambling and Last Resort". Linguistic Inquiry 29.347-366. Chomsky, Noam. 1993. "A Minimalist Program For Linguistic Theory". The View from Building 20 ed. by Kenneth Hale and Samuel J. Keyser, Cambridge, Mass.: MIT Press. Reprinted in N. Chomsky (1995). Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. Chomsky, Noam. 1998. Minimalist Inquiries: The Framework. MIT Occasional Papers in Linguistics 15. Reprinted in Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik ed. by Roger Martin, David Michaels, and Juan Uriagereka. 2000. Cambridge, Mass.: MIT Press. Chomsky, Noam. 1999. Derivation by phase. MIT Occasional Papers in Linguistics 18. Galves, Charlotte. 1992. "Inflected Infinitive and Agr Licensing". Ms. University of Campinas (Unicamp), São Paulo. Jaeggli, Osvaldo A. 1986. "Arbitrary Plural Pronominals". Natural Language and Linguistic Theory 4.43-76. 1999. "Movement and control". Linguistic Hornstein, Norbert. Inquiry 30.69-96. Hornstein, Norbert & Itziar San Martin. 2000. "Obviation as anti-control." Paper presented at GLOW 23, Main Session. Vitoria-Gazteiz, Spain. Kempchinsky, Paula M. 1986. Romance Subjunctive Clauses and Logical Form. Ph.D. Dissertation, UCLA. Lasnik, Howard. 1995. "Last Resort and Attract F". Proceedings of FLSM 6, 62-81. Indiana University Linguistics Club, Bloomington. Lebeaux, David. 1985. "Locality and Anaphoric Binding". The Linguistic Review 4.343-363. Martin, Roger. 1996. A Minimalist Theory of PRO and Control. Ph.D. dissertation, University of Connecticut, Storrs. Modesto, Marcelo. 1999. "Null Subjects without Rich Agreement". Ms. USC, Los Angeles, CA. Nunes, Jairo. 1995. The Copy Theory of Movement and Linearization of Chains in the Minimalist Program. Ph.D. dissertation, University of Maryland, College Park.
PRO, MOVEMENT AND BINDING IN PORTUGUESE
167
Pires, Acrisio. Forthcoming. "Infinitives, Control as Movement and the Loss of Inflection in Portuguese". Syntactic Effects of Morphological Change ed. by David Lightfoot. Oxford: Oxford University Press. Raposo, Eduardo. 1987. "Case Theory And Infl-to-Comp: The Inflected Infinitive In European Portuguese". Linguistic Inquiry 18.85-109. Rodrigues, Cilene. Forthcoming. "Loss of Verbal Morphology and the Status of Referential Null Subjects in Brazilian Portuguese". Syntactic Effects of Morphological Change ed. by David Lightfoot. Oxford: Oxford University Press. San Martin, Itziar. 1999. "Control in Basque". Doctoral Research Paper, University of Maryland. To appear in University of Maryland Working Papers in Linguistics. Uriagereka, Juan. 1998. Rhyme and Reason: an Introduction to Minimalist Syntax. Cambridge, Mass.: MIT Press.
ON IMPERSONAL REFLEXIVES IN ROMANCE AND SLAVIC AND SEMANTIC VARIATION * MARIA LUISA RIVERO University of Ottawa
0.
Introduction This paper deals with impersonal se/si/się in Romance and Slavic. It shows that in its impersonal use, the reflexive clitic of Spanish, Italian, and Polish displays identical properties, except for a striking difference that separates the Slavic language from Romance. It is argued that this contrast is due to parametric variation in semantics, which affects the impersonal as indefinite pronoun. This proposal should be of interest for linguistic theory for several reasons. One is that parametric variation has received less attention in semantics than in syntax, and needs to be explored. Another one is that the functional categories are the traditional locus of syntactic variation, and it seems that semantic variation may also reside in them, in particular in clitics such as se/si/się. Romance and Slavic reflexive clitics share many uses illustrated in (1) with Spanish and Polish. All languages display the uses often called (a) reflexive/reciprocal, (b) middle/passive, (c) anticausative/inchoative/
* This paper has been partially supported by SSHRCC Research Grant 410-97-0242. Partial preliminary versions were presented at Going Romance in Utrecht in 1998, the Fundación Ortega y Gasset in Madrid and the Glow Summer School in Mytilene in 1999, and LSRL 30 and the Universities of Brasilia and Campinas in 2000. I am grateful to the various audiences for useful comments, and most thankful to the organizers of LSRL 30 for inviting me, which gave me the nostalgic opportunity to return to the site of the first Romance Symposium, where I also participated (Rivero 1972). Special thanks go to G. Chierchia for reasons that will be evident by section 3. M. Sheppard and I are researching this topic in Slavic (Rivero & Sheppard 1999), and I thank her for discussion and for information on Slovenian. I owe many thanks to A. Boron, B. Borsley, M. Goledzinowska, E. Jaworska, A. Przepiórkowski, and E. Willim for Polish, O. Arnaudova for Bulgarian, and C. Diaconescu for Rumanian.
MARIA LUISA RIVERO
170
unaccusative/ergative, and (d) inherent/intrinsic. Unless otherwise indicated, from now on, examples are first in Spanish followed by Polish. (1)
a. a'
b. b'
c'
d. d'
Juan se viste. John Refl dresses Janek ubiera się. John dresses Refl "John gets dressed." Este coche se conduce fácilmente. this car Refl drives easily Ten samochód powadzi się 1atwo. this car drives Refl easily "This car drives easily." El vaso se rompió. the glass Refl broke Szklanka się rozbiia. glass Refl broke "The glass broke." María se asusta de Juan. Mary Refl fears of John Maria hoi się Janka. Mary fears Refl John "Mary {fears / is afraid of} John."
In Romance, examples similar to (1) could be given in at least French, Italian, Portuguese, and Rumanian. All Slavic languages with reflexive clitics also share these uses (besides Polish, Bulgarian, Croatian / Serbian, Czech, Macedonian, Slovenian, and Slovak). With the exception of French, all the mentioned languages display the use in (2) with Vs that traditional grammars call intransitive. This use, which serves for the two types of intransitives distinguished in generative grammar (i.e. unergatives (work), and unaccusatives (die)) has been called impersonal, intransitive, {unaccusative/ergative}, or unergative. (2)
a. b.
Aquí here Tutaj here "Here
se Refl się Refl people
trabaja works pracuje works work a lot."
mucho. much sporo. much
ON IMPERSONAL REFLEXIVES IN ROMANCE AND SLAVIC
171
It has been argued (Dobrovie-Sorin 1998) that intransitive patterns with reflexive clitics are open to two analyses. In one analysis, they formally resemble the passive/middle patterns above. In the other analysis restricted to the languages with the patterns in (3), they contain what is called in this paper the Nominative Impersonal. A much debated use glossed with "one" or "people" and dubbed here Nominative Impersonal (also called indefinite, indeterminate, nominative, nominativeless, or subjective) is the topic of this paper, and separates Romance and Slavic into two groups, as in Table 1. In Romance, this use is absent in French and Rumanian, and present in Italian, Portuguese, and Spanish, where it consistently cooccurs with a 3S predicate: (3a) among other cases. In Slavic, it is found in Slovenian and Polish, where it cooccurs with a 3S predicate in the Present, and a NEU(ter) predicate in the Past: (3b). Yes ROM: Italian; Portuguese; Spanish. SLA: Polish; Slovenian.
No French; Rumanian. Bulgarian; Czech; Slovak; Serbo-Croatian (?).
Table 1 : The nominative impersonal (3)
a.
b.
Antes se leía estos libros con placer. before Refl read.3S these books with pleasure "In the past {one/people} read these books with pleasure." Tęksi^zkq {czyta/czytalo} się z przyjemnościa. this book.ACC {read.3S/read.NEU} Refl with pleasure "One {reads/read} this book with pleasure."
In dealing with the use in Table 1, this paper develops the syntactic analysis in Rivero (1999) first proposed for Polish, and applies the semantic proposal in (Chierchia 1995) to Spanish and Polish (see also Rivero and Sheppard 1999 for Slavic). This global treatment captures form and meaning characteristics shared by the Romance and Slavic impersonal. In the Yeslanguages of Table 1, this analysis also serves for intransitive patterns of type (2). The core syntactic idea in section 2.1 is that the impersonal construction is defective in two ways. One, the reflexive clitic signals a defective pronoun or syntactic S(implex) E(xpression) anaphor in the sense of (R(einhart) & R(euland) 1993). Two, in the impersonal construction finite T(ense) is defective. In a few words, clitic se/si/się cooccurs with a null NP, not a DP (Rivero 1999), which is the external argument of the predicate and is located in the Spec of the VP. This NP has a human feature and no PHI-features, so
MARIA LUISA RIVERO
172
cannot be interpreted independently (-R). Since T is defective in that it lacks PHI-features, the NP must raise to the 'base-generated' clitic to repair its (a) formal and (b) referential deficiency. Movement allows the NP to (a) check Case (NOM(inative)) against the clitic, which removes its uininterpretable feature as formal imperfection, and (b) acquire existential force, which repairs its referential imperfection. More technically, the NP is - R and Nominative, and by raising to the clitic, which also has Case, it becomes the member of a chain with a +R head and a - R tail in the sense of R&R. The impersonal can thus be interpreted as a semantic indefinite pronoun in the sense of Chierchia (1995), as discussed in section 2.2. In sum, the impersonal in (3) and similar cases involves the syntactic chain of a defective human pronoun - a SE-anaphor - and the semantics of an indefinite with intrinsic existential force. As stated, this analysis covers the intransitive constructions with reflexives of type (2) in the Yes-languages of Table 1, but is not for intransitive constructions in the other languages. The major difference between the two languages in Table 1 then is that the reflexive clitic may stand for a defective nominative human indefinite pronoun only in the first group. In section 3, it is shown how this indefinite pronoun analysis can account for the contrast in (3) vs. (4). The Romance and Polish impersonals display almost identical properties, but differ in a striking way with unselected Datives. (4)
a.
b.
Antes se me leía estos libros con placer. before Refl me.DAT read.3S these books with pleasure "Before {one/people} read these books to me with pleasure." Tęksi^zkq czytalo mi siq zprzyjemnosci^ . this book.ACC read.NEUT me.DAT Refl with pleasure Dominant reading: "I (AGENT) read this book with pleasure." Dispreferred but possible : "{One/people} read this book to me with pleasure."
In Spanish (4a), the unselected (Ethical) Dative has no effect on se, which translates as "one" or "people", as in (3a). Polish shares with Spanish the Ethical use of Datives, but in (4b) Dative mi "me" is preferentially interpreted as Agent, which altogether suppresses the impersonal reading in (3b) (and see Dziwirek (1994) for discussion). The Spanish and Polish impersonals are very similar, so the difference in (4) is at first sight unexpected. Rivero (1999) suggests that siq as impersonal can play an expletive role that is not
ON IMPERSONAL REFLEXIVES IN ROMANCE AND SLAVIC
173
shared by Romance impersonal se/si. In section 3 it is argued that this role is the result of parametric variation in semantics. The impersonal is an existentially quantified indefinite pronoun optionally disclosed by adverbs of quantification that come to bind it, which applies to Italian, Spanish, and Polish. What distinguishes Polish is that the impersonal may also be disclosed by an (unselected) Dative that lends it its content. On this view, the explanation for the contrast in (4a) vs. (4b) is semantic. It arises from the impersonal use of the reflexive clitic, a functional category, coupled with the Ethical Dative. It consists of a specific choice of discloser in Polish. In other syntactic and semantic respects, the impersonal seems very similar in Romance and Polish. 1. Syntactic diagnostics in the Impersonal This section develops ideas in Rivero (1999), outlining some familiar and some unfamiliar syntactic properties of the Impersonal in Table 1 that are important to view it as a defective pronoun or SE-anaphor, as in section 2.1. Section 2.2 applies the semantic proposal of Chierchia (1995) to Spanish and Polish, adding other meaning characteristics and setting the stage for the explanation of the contrast in (4). First, two diagnostics of the impersonal construction much debated in generative grammar since Chomsky (1981) concern case and agreement, and play a part in the present account. Beginning with case, a common assumption is that impersonal se/si/siq requires NOM, and a current implementation of this idea will be given in 2.1. That NOM is somehow tied to the impersonal receives support from several phenomena. These include (a) the morphological ACC(usative) on the overt NP in Polish (3b), viewed by many as an indication that the construction is 'active' and not 'passive', (b) the preposition a preceding the overt NP in Spanish (5a), and (c) se/się combined with an ACC clitic in (5b-c)." (5)
a.
b. c.
En esta escuela se castiga a los alumnos. in this school Refl punish.3S a the students "In this school one punishes the students." Si una niña es mala, se la castiga. little.girl is bad, Refl her punish.3S if a Jeśli dziewczynka jest niegrzecsna, haze się ja. if little.girl is bad punish.3S Refl her "If a little girl is bad, one punishes her."
MARIA LUISA RIVERO
174
Another factor favoring NOM in (Rivero 1999) is that only those Polish Modals that accept NOM subjects may cooccur with the impersonal. The last factor is a difference in Tough-constructions, which I think has gone unnoticed. In Romance, the overt NP is NOM, as in (6a), so impersonal versions exist for this construction, as in (6b). (6)
a.
b.
Ellos y no ellas son difíciles de contentar. they.NOM.and not they.NOM are difficult of please.INF "The males and not the females are difficult to please." A quindici anni, si è difficile da accontentare. Italian at fifteen years, Refl is difficult of please.INF "At fifteen, one is difficult to please." (Chierchia 1995)
In Polish, there is no exact equivalent for (6a) with a NOM NP. One grammatical option reminiscent of English It is difficult to please John is as in (7a). It has an initial ACC NP scrambled out of the embedded clause that bears the Case required by the lower V. Due to this difference arising from Case, Polish differs from Romance in having no się -impersonal Toughconstruction. That is, (7b) is deviant. (7)
a.
b.
Jana trudno jest John.ACC difficult is "John is difficult to please." *Trudno się jest zadowolic.
zadowolic. please.INF
As to agreement, the Romance V is 3S, as in (3a), (4a), and (5a-b). The Slavic V is 3S in the Present - (3b) and (5c)-, and NEU in the Past: (3b-4b). Thus, the tensed item does not agree in Person/Number/Gender with the overt NP, which bears the Theta-role usually assigned to objects, and in Slavic Accusative Case in affirmative clauses, and Genitive Case in negative clauses (Rivero 1999). Case/agreement in (3-5) are morphological clues to distinguish between impersonal and most notably {passive/middle} constructions. The last display a NOM NP, shown morphologically in Polish, and a tensed predicate that agrees with it in PHI-features, as in (lb-b'). We know that several Romance and Slavic languages lack the Nominative Impersonal. That is, they do not have a reflexive clitic treated as a defective indefinite human pronoun. These languages fail to display the
ON IMPERSONAL REFLEXIVES IN ROMANCE AND SLAVIC
175
case and/or agreement patterns in (3-5), as in Rumanian (8a) and Bulgarian (8b) for (5a). (8)
a.
*In s ala asta se pedepseşte pe elevi. in school this Refl punish.3S pe students *V tova učilište učenizite se nakazva. in this school students.the Refl punish.3S
Rumanian Bulgarian
A second diagnostic of se/si/się. in Table 1 has attracted less attention: it can bind or antecede all types of anaphors. If binders must be present in the syntax, then se/si/się. signals a syntactically projected NP in an A-position that binds the anaphor. Italian supports the hypothesis that the impersonal can bind a clitic anaphor. In (9), si is the impersonal "people", and ci the reflexive anaphor "themselves" it binds (since the impersonal follows object clitics in Italian). (9)
Ci si è lavati. Self-cl Refl be.3S washed.PL "People washed themselves." (Chierchia 1995)
Italian
Spanish and Polish prohibit *se se/*się się, as noted by Perlmutter (1971) for Spanish. No phonological process changes the shape of those sequences, so the option in (9) is unavailable. However, the impersonal can bind non clitic local anaphors, such as uno mismo and sobie in (10). (10) a. b.
Ahora Teraz now "Now
se piensa sólo en {uno mismo /?si mismo}. siq myśli tylko o sobie. Refl think.3S only {in/of} oneself one thinks only of oneself."
The Slavic languages have (long distance) possessive anaphors, and (11) illustrates that the Polish impersonal can antecede this type. (11) Swoich przyjaciól tak siq nie traktuje. POSS friends so Refl Neg treat.3S "One does not treat one's friends like that." (Siewierska 1988)
MARIA LUISA RIVERO
176
Italian and Polish have intersentential long distance anaphors like proprio and the possessives. The sentences in (12) illustrate that the impersonal can also antecede this type. (12) a.
b.
Si ritiene (spesso) che i proprio errori siano pin ' giustificati Refl believe.3S (often) that the own errors are more justified di quelli degli altri. Italian than those of.the others (Chierchia 1995) Myśli się, ze swoje b1ędy sa bardziej believe.3S Refl that POSS errors are more usprawiedliwiane niz innych. justified than of. others "People (often) think that their own mistakes are more justified than those of the others."
Sentences (11) and (12) seem particularly significant for the idea that the impersonal is an (independent) syntactic argument, and not the morphological reflection of a predicate that in the lexicon is inherently reflexive. These two examples involve non-local anaphors, which cannot reflexivize the predicates "treat" and "think" in the sense of R&R's theory (1993). Since these predicates are not semantically reflexive, si and się cannot be (redundant) markers of intrinsic reflexivity, which is the analysis often suggested for reflexive clitics in the Romance languages. These examples show the need for two positions in the syntactic structure: (a) one for the impersonal, which is not a reflexivizer nor a marker of lexical reflexivity, and (b) another one for the long distance anaphor. If the impersonal was seen as only a marker of intrinsic reflexivity, there could be only one argument in sentences like (l0a-b), namely the constituent uno mismo/sobie. However, if (11-12) must contain two argument positions, this option must also be available in (10); thus, here too the impersonal is the binder, and the non-clitic anaphor is what R&R call a SELF-anaphor. In other words, in (10) uno mismo and sobie are extrinsic markers of reflexivity that reflexivize the predicate "think-of"', which is not intrinsically reflexive. The Italian impersonal is plural (see below). This allows it to antecede a reciprocal (Belletti 1982), which is excluded in most of the languages in Table 1 (but apparently not in Slovenian, as M. Sheppard points out to me).
ON IMPERSONAL REFLEXIVES IN ROMANCE AND SLAVIC
(13) Si era paríalo l'uno con Refl was spoken the one with "{They, we} had spoken to each other."
I'altro. the other
177
Italian
The fact in (13) is also significant under the reflexivity theory of binding. On this view, the impersonal serving as antecedent for a reciprocal cannot be a marker of intrinsic or extrinsic reflexivity. This type of example signals more than one syntactic position in the construction, irrespective of how reciprocals, which are not covered by R&R's theory (1993), are treated. The Indoeuropeanist tradition relates uses of the reflexive to intransitivity (Schenker 1988 for earlier views on Latin, Romance, and Slavic). In the Romance and the Slavic generative literature, several attempts to unify the many uses of se/si/się rely on some form of reflexivity: the coindexation of two argument positions in the syntax, or the suppression of one of them via intransitivization (Cinque (1988) for references on Romance, and Rivero (1999) for references on Polish). However, the facts just reviewed suggest that reflexivity/intransitivity is unlikely to account for the impersonal, which intuitively does not behave like an anaphor nor an intransitivizer. The proposal in section 2.1 is that the impersonal belongs with defective pronouns. In languages without the impersonal, se is not a binder or antecedent. To illustrate, Bulgarian has possessive anaphors like Polish, but lacks impersonal se. The Bulgarian (passive-middle) se cannot function as the (needed) antecedent, so (14) is deviant and not "One sees one's pictures with pleasure." (14) *Svoite kartini POSS pictures
se Refl
gledat s udovolstvie. Bulgarian see.3P with pleasure
The third characteristic of the impersonal interesting for our purposes concerns Control. It is well known that the impersonal can control into adjuncts, and less known that it can control into passive complements, as in (15). This diagnostic is significant under the assumption that control into passives must be by an argument present in the syntax (Jaeggli 1986). On this view, the impersonal signals the syntactically projected controller. Given what was said about Case, the controller is NOM, and given binding, it is in the equivalent of a syntactic A-position.
MARIA LUISA RIVERO
178 (15) a. b.
Siempre se quiere ser {admirado/apreciado}. always Refl want.3S be {admired/ appreciated} Chce się być {admirowanym /lubianym}. want.3S Refl be {admired/liked} "One (always) wants to be {admired/liked}."
The fourth characteristic that plays a role in our analysis has also been noted by many in the past. Similar to arbitrary PRO, impersonal se/si/się must denote a human or sentient being such as a personified animal. In an interesting semantic study of the Polish reflexive, Kański (1986, 1992) assigns human aspect to pragmatics, but, here we agree with most in locating it in the semantics of impersonal se/si/się (in particular Chierchia 1995). In our view, the only interpretable feature in the impersonal is [+Human/Sentient]. This, coupled to the idea that it is an indefinite, places Spanish se in the paradigm of quantificational-like intrinsically human lexical entries such as alguien "someone'1, nadie "no one", and quien/quienes "who". The Spanish verb amanecer "to dawn" is useful to highlight human aspect, and to show that in the Yes-languages of Table 1, the Nominative Impersonal analysis extends to intransitive Vs (for the languages without impersonal see Dobrovie-Sorin (1998)). This V need not involve humans/animates, so a common reading for (16a) refers to just the beginning of the day. However, when impersonal se is added, as in (16b), reference must be to {humans/personified animals}. This sentence is, for instance, appropriate when speaking to a budgie that chirps happily early in the day. (16) a.
b.
Hoy amanece bien. today dawn.3S well "The day begins well." Hoy se amanece bien. today Refl dawn.3S well "Today {people/you/we} {wake up / begin the day} well."
In section 3, the {human/sentient} feature acquires additional formal significance when it is argued that the disclosure of the impersonal by the unselected Dative in Polish requires that this feature be matched. A last characteristic makes the impersonal resemble an ordinary pronoun or indefinite and is that it can usually be assigned or tied to the same semantic (Theta) roles as an ordinary (overt) Nom subject. Among the properties
ON IMPERSONAL REFLEXIVES IN ROMANCE AND SLAVIC
179
highlighted here, we see next that this is the less general. However, its exceptions are interesting for the general semantic analysis of the impersonal. First, it is well known that in finite clauses1 the impersonal in Italian, Spanish, and Polish can combine with all types of Vs, including modals, passives, and the copula constructions with adjectives illustrated in (17-18). (17) Se si è alti, si è talvolta anche belli. if Refl be.3S tall.PL, Refl be.3S often also handsome.PL "If one is tall, one is often also handsome." (18) a. Cuando se es joven, raramente se es feliz. b. Kiedy się jest mlodym, rzadko się jest szczę śliwym. when Refl be.3S young, seldom Refl be.3S happy "When one is young, one is seldom happy." Copulas with adjectives or participles (i.e. passives) prove significant from several perspectives, and I mention two. One, an important number of adjectives are individual level predicates, as in the above examples. Thus, when this type of adjective coexists with impersonal se/si/się, semantically it signals an individual variable in the construction. This fits well with the clues given above indicating that the impersonal is a syntactically projected NP that as argument with a human feature can bind, antecede, and control. I return to this clue immediately below, since there are exceptions to it. Two, constructions with adjectives or passive participles show a principled morphological connection between the impersonal and arbitrary PRO. The (Present) copula is always 3S in all the languages, but adjectives (or participles) vary to consistently display in each language the morphology also required by arbitrary PRO. In Italian, this PRO is masculine/plural, which is the morphological marking of the adjectives in (17). In Spanish, it is most often masculine/singular, as with the adjectives in (18a). In Polish, PRO usually cooccurs with a masculine adjective in the Instrumental, which is found in (18b) as well; here NOM would be deviant. The impersonal seems devoid of intrinsic PHl-features, so these agreement patterns are the default setting, and others are possible. Concerning Gender, Spanish (or Polish) feminine adjectives are appropriate when a woman is either speaking or spoken to; (19a), for instance, could be used by a male doctor commenting on 1 I omit discussion of nonfmite clauses. In Romance, except for Portuguese, the impersonal is not found in nonfinite clauses (Dobrovie-Sorin (1998) on Italian), or is very restricted (Mendikoetxea (1999) on Spanish). In Slavic, it is not found in such clauses (Ruzička 1992).
MARIA LUISA RIVERO
180
the fair skin of a female patient. Concerning Number, the Spanish impersonal is not incompatible with plurality, as in (19b); similar examples are possible in Polish. (19) a.
b.
Cuando se es tan blanca, el sol es peligroso. when Refl be.3S so white.FEM, the sun is dangerous "When a woman is so white, the sun is dangerous." Aquí se entra de dos en dos. here Refl go.3S.in of two in two "Here people go in two by two."
Copula/adjective constructions as in (17-18-19a) and passives are not found with the impersonal in all the languages of Table 1. In Slavic, Slovenian impersonal se shares the properties above (case/agreement, binding, control, human denotation), but cannot combine with copulas + adjectives or passives (Rivero and Sheppard 1999). Thus Slovenian (20) is deviant, making this language resemble Bulgarian and Rumanian, two of the languages without the impersonal of Table 1 (that is, languages that have no se that counts as a defective human pronoun in the sense of section 2). (20) *Ko se je mlad, se je srecen. when Refl be.3S young Refl be.3S happy Intended: "*When one is young, one is happy."
Slovenian
Diachronic evidence points to the same situation in Romance. Nunes (1991) tells us that the grammarian Said reports that Portuguese passives and copulas combined late with impersonal se, and are found in the literary language only in the 19th century. Thus, for a time Portuguese and Slovenian must have been similar. A detailed analysis of this restriction is beyond the scope of the present paper, but I sketch a proposal in informal terms. The Slovenian or the earlier Portuguese impersonals display the diagnostics of a syntactically projected NP with all the relevant properties. Thus, they must also correspond to the human indefinite pronoun discussed in section 2; we shall see that this pronoun is a SE-anaphor, and contains an existential quantifier and an individual variable. The restriction seems to be that the Slovenian and earlier Portuguese individual variable in the impersonal use of the reflexive clitic must be licensed by being coupled to a so-called {situation / event} variable. On this view, impersonal sentences with transitive Vs in the active voice,
ON IMPERSONAL REFLEXIVES IN ROMANCE AND S L A V I C 1 8 1 those with unergative Vs such as "work", and those with unaccusative Vs such as "arrive" all contain a {situation/event} variable needed to license the impersonal as indefinite with its individual variable in Slovenian and earlier Portuguese. Thus the impersonal can associate with an external Th-role or an internal one, without conflict. However, in constructions of a stative nature, as with the copula and adjectives and passive participles, the situation variable is either absent or inaccessible, and the Slovenian and earlier Portuguese impersonal is rendered illicit. From this perspective, when the impersonal participates in the four symptomatic properties discussed above, it must correspond to an indefinite pronoun with a human individual variable that is syntactically projected. However, in some cases, this variable needs to be supported by a situation variable.2 In sum, the syntactic diagnostics of the impersonal as defective human indefinite pronoun highlighted here as most significant for the analysis in section 2 are the following: I. The impersonal relies on a functional category with the same clitic characteristics as other non-tonic pronouns. II. It may coexist with an ACC NP, or ACC clitic. In Polish it may appear with a Modal that takes NOM subjects, but not in Tough constructions. Thus, it is related to NOM. III. It coexists with a tensed predicate without PHI-features, which indicates that T is defective. IV. It can be a binder or antecedent for many types of anaphors, signaling what amounts to an argument in A-position in the syntax. Long distance anaphors, and reciprocals are of particular importance within reflexivity theory. This is because they indicate two syntactically projected positions, one related to the impersonal (as antecedent), and one related to the long distance anaphor or the reciprocal. 2
The proposals on Italian infinitives with si in Cinque (1988) could be used as a basis to suggest that the Slovenian and (earlier) Portuguese restriction indicates a need for an external Theta-role (Dobrovie-Sorin (1998) for counterproposals). However, I just rejected this option given that Slovenian and earlier Portuguese modals, which may pattern with raising Vs, and unaccusative Vs, not associated with an external Theta-role, easily combine(d) with se. In Portuguese se is found with copula + adjectives and passives late and with unaccusatives early. Nunes (1991: 45) documents it with chegar "arrive" in 1677. The distinction between individual-level and stage-level predicates does not seem appropriate to capture the Slovenian and earlier Portuguese restriction. Predicates like "intelligent" and "love/hate" are both considered individual-level, but the Slovenian impersonal is problematic with all adjectives and fine with "love/hate". The distinction (a) between non-states and states, (b) "love/hate" as activity Vs or non-states with a situation variable, and (c) adjectives as statives without this variable seem to capture the difference.
MARIA LUISA RIVERO
182
V. It also signals a type of controller that must be syntactically projected and not implicit. VI. It has no PHI-features of its own, but involves formally encoded {human / sentient} denotation, with possible morphological consequences, as with agreement and case on passive participles and predicative adjectives. VII. It can be associated with a variety of external and internal Thetaroles held by an (overt) Nom subject. Thus, it is used with many types of predicates to the same effect as full-fledged (NOM) pronouns and indefinites. Copula + Adjective constructions seem of particular significance for the syntax - semantics interface of the impersonal. Adjectives that are individuallevel predicates signal an individual variable when in construction with the impersonal. The individual variable diagnostic fits well with criteria I-VI, which indicate on other grounds that the impersonal is a syntactically projected argument NP with NOM case, a human feature, and can bind, antecede, and control. The Slovenian and earlier Portuguese impersonal use of the reflexive clitic shares the diagnostics in I-VI, but cannot appear in copula + adjective and passive constructions. In these languages or periods, the impersonal is also a defective indefinite pronoun that contains a quantifier and an individual variable, and is unrestricted as to Theta-role, but can only be licensed with the help of a situation variable. Romance languages like Rumanian and Slavic languages like Bulgarian do not display these diagnostics because their reflexive clitic is not analyzed as a human defective pronoun, which includes intransitive constructions. 2. The analysis 2.1 The impersonal as Simplex Expression Anaphor. Rivero (1999) argues that the properties of the impersonal in section 1 can be captured by the assumption that it is a S(implex) E(xpression) anaphor, or defective pronoun, and in this section I summarize this proposal. The idea is implemented by taking the clitic to be {"base generated'Vmerged} outside the VP, attracting a defective (null) NP that is the external argument of V, so in the equivalent of Spec-of-VP. This movement hypothesis is coupled with (the spirit of) the binding theory of R & R (1993). On this analysis, the skeleton assumed for the impersonal sentences in (3) is as in (21): (21)
[CLP
[c1 se/si/się] [TP [ T Pres / Past] [VPNPI
V
NP2]]]
ON IMPERSONAL REFLEXIVES IN ROMANCE AND SLAVIC
183
V heads a VP that contains two NP arguments. The "arbitrary" NP1 as external argument is equivalent to a null defective pronoun, different from little pro: it has a human feature, (structural) NOM Case, but no PHl-features. If this null expression constitutes a syntactic phase, then it should be considered a NP and not just a N (Rivero (1999) for reasons not to see it as a DP), which fits well with the idea that it is a semantic indefinite. The internal argument NP2 is the overt ACC object this/these book(s). The T(ense) P(hrase) is headed by T(ense), which is defective in that it also lacks PHIfeatures. TP takes VP as complement and V checks features against T. Given that T is defective, V is either 3S or NEU, i.e. without PHI-features. The other functional projection dubbed Cl(itic) P(hrase) is headed by se/si/się. The core idea is that in (21), NP1, which is a syntactically projected argument of the predicate for the reasons stated in section 1, must repair deficiencies by raising to (the Specifier of) C1P, checking structural Case. Thus, NP1 in the Impersonal construction contrasts with the null pronoun little pro of the ordinary null subject sentence, which is often assumed to check features with a non-defective T in TP. The clitic is a functional category directly merged outside of the VP (known as the "base generation" approach), and NP1 raises to it, so let us see why. Chomsky (1998, 1999) proposes two operations to check formal features: MOVE and AGREE, which is more economic. Oversimplifying, a category can MOVE to the Spec of a functional head to check features, or the functional head can check features with a category in its complement via AGREE, without the need for movement. The last operation is preferred. One important question in this framework, then, is why MOVE and not AGREE applies in the impersonal; what forces the deficient NP in (21) to leave the VP. Another important question is why MOVE must target functional se/si/ się. Structural Case combined with the lack of PHI-features can provide a formal answer to these questions. In (21), both NP1 and T lack PHI-features. Following Chomsky, this lack of PHI-features signifies that these two categories cannot establish an appropriate checking relation with each other. Thus, I propose that the required relation is established between NP1 and the clitic, when the first moves to the second. These categories are each equipped with a structural Case feature, which allows them to match for checking (for further discussion see Rivero (1999)). In brief, absence of PHI-features on both the clitic and the verb, two prominent characteristics of the impersonal construction, makes the NP raise to CL to satisfy formal needs (feature checking for Case).
184
MARIA LUISA RIVERO
Now consider binding, the other dimension in the analysis. R&R (1993: section 1) distinguish between Pronouns, SELF-anaphors, and SE-anaphors, with two features: Refl(exivizing function) and R(eferential Independence). Pronouns such as him in John hates him are not reflexivizers and contain PHIfeatures, which allows them to be interpreted independently: [-Refl; +R]. SELF-anaphors such as himself in John hates himself are referentially defective, and reflexivizers [+Refl; - R ] . SE-anaphors are like pronouns in not being reflexivizers, and like SELF-anaphors in not having a full specification of PHl-features, so cannot be interpreted independently: [-Refl; - R ] . The content necessary for their interpretation is obtained via movement, which makes them similar in feature specification to a pronoun: [+R; -Refl]. Dutch zich is a SE-anaphor. It does not make a predicate reflexive when in one of its argument positions: *Max haat zich / Max hates SE-anaphor. It lacks PHIfeatures, so it cannot be interpreted independently. However, it can obtain the feature specification necessary for its interpretation by movement. In Max legt het boek achter zich "Max puts the book behind him=Max", zich adjoins in LF to I ( = T) to inherit subject features (1993:659), which results in a well formed A-chain that is +R and Case-Marked: Max zich-legt het boek achter t. The movement does not make the predicate put reflexive, but coindexes zich with Max as subject. In sum, for R&R a SE-anaphor is a defective pronoun that repairs deficiency by acquiring PHI-features via a LF movement that coindexes it with the subject. The impersonal use of se/si/siq is comparable to a Dutch SE-anaphor in three ways, but there are interesting differences. One, since the null NP in (21) lacks PHI-features, it does not project an argument that can be interpreted independently. Two, movement enables this NP to obtain the content for its interpretation. R&R's SE-anaphors adjoin to I to inherit the PHI-features of the subject, which makes them (a) "subject-oriented", and (b) interpretable as ordinary personal pronouns with a full set of PHI-features. However, our defective NP repairs its deficiency by moving to CliticP with se/si/siq, which lacks PHI-features. The movement in the impersonal construction is thus different from the movement in the Dutch sentence above, but achieves the same syntactic result, which is to create a well-formed LF-chain that can serve as input for semantic interpretation. For R&R, an A-chain is well formed if it is +R and Case-Marked. In our case, the movement of the PHIless NP up to the clitic ensures that it can be considered +R, which allows it to be interpreted in the way to be discussed in 2.2. Since the NP checks Case, the chain is also Case-Marked. The similarity between the SE-anaphors involving a clitic as target and the Dutch SE-anaphors, which need not
ON IMPERSONAL REFLEXIVES IN ROMANCE AND SLAVIC
185
involve a clitic, is that movement of a PHI-less NP in an argument position of a predicate results in a chain that is interpretable at LF. The third aspect that makes the impersonal like a SE-anaphor is the Binding Theory, highlighting its pronominal character: impersonal se/si/się is a clear pronoun. SE-anaphors pattern with pronouns in that they occur in an argument position of a predicate without making this predicate reflexive. On this view, the defective null NP that serves as double for the clitic in the impersonal use patterns like (Chomsky a pronoun. It fits in unproblematic ways traditional principle 1981): a pronoun is free in its governing category. Alternatively, following R&R, the impersonal does not make the predicate reflexive when it occurs in one of its argument position (i.e. if used impersonally the reflexive clitic corresponds to "one, people"). R&R propose no principle to prevent a SEanaphor from being free, and we saw above that Dutch zich is coindexed with the subject not because of the binding theory, but due to the movement for PHI-features. The impersonal uses of se/si/się behave like pronouns rather clearly. Intuitively speaking, they are "antecedentless" expressions. These uses, then, are pronominal as they indicate a SE-anaphor, which does not reflexivize the predicate (-Refl), and repairs referential deficiency by moving to se/si/się (in R&R's terms, the null NP changes its value from [-R] to [+R] when it raises to the clitic). In sum, se/si/siq with a NOM feature is merged in CL. There is a NP in the sentence without PHI-features, so it cannot enter into an AGREE relation with T, which also lacks PHI-features. The NP checks Case by MOVING to se/si/siq. This operation results in a LF-chain interpreted as an expression with a pronominal and a human character, which is the impersonal as SEanaphor. 2.2 The impersonal as indefinite pronoun (Chierchia 1995) In the analysis in 2.1. the NP in (21) moves to the "base-generated" clitic. This results in a LF chain that is well-formed, providing an appropriate input for later levels of semantic interpretation. What are the semantics of the impersonal? Chierchia (1995) argues that Italian impersonal si is an indefinite pronoun. This hypothesis is compatible with our syntactic proposals, can serve for Spanish and Polish (and see Rivero and Sheppard (1999)) for more details, and discussion of Slovenian), and captures important interpretive parallelisms shared by the Romance and Slavic impersonal. In my view, the Polish and Spanish impersonals as indefinite pronouns do not differ from their Italian counterpart. Thus, in this section I illustrate the gist of Chierchia's proposal via these languages, concentrating on aspects that
MARIA LUISA RIVERO
186
play a role in section 3 (the reader is referred to Chierchia's work for the fullfledged proposal). Four aspects in the semantics of the impersonal will help us understand in section 3 the role of unselected Datives in Polish. One, the impersonal is an indefinite pronoun with a sortal restriction (i.e. human). This fits well with the proposal in 2.1 that it is a SE-anaphor with a formally encoded human feature. Two, in the line of Frege, Russell, and dynamic semantics, Chierchia takes indefinites to have intrinsic existential force. By contrast, in her wellknown proposal, Heim (1982) sees them as free variables without quantificational force. The idea that indefinites have force also fits well with the syntax in 2.1, suggesting why a clitic is needed in the impersonal reading. The movement in 2.1 forms a chain whose head combines the raised human NP and a clitic equivalent to an operator that provides force, and the foot contains a trace (or copy) that counts as a variable. Thus, the syntactic chain shares A and A-bar characteristics. On the one hand, it is a pronoun of the SE-anaphor type, or an A-chain involving the Case system and formal feature-checking, as in traditional NP-movement. On the other hand, it is an A-bar chain or an indefinite with existential force, involving variable binding, as in traditional Quantifier Raising. Three, while indefinites are intrinsically existential in this analysis, they may be disclosed by adverbs of quantification that come to bind them, so show quantificational variability. Four, the disclosure procedure is optional, so the impersonal may preserve its own force in the contexts where it could be subject to variability. By contrast, for Heim (1982), indefinites must get their force from adverbs of quantification, or undergo existential closure by default at the text level. Let us now briefly illustrate each point, giving partial motivation. The impersonal is a pronoun, so it can antecede itself, as in (22). Thus, it escapes Heim's Novelty Condition for indefinites, which covers cases like If a player does not play well, {he/*a player} loses. For Chierchia, the Novelty Condition is not independent, but a consequence of principle for NPs (Chomsky 1981), which does not regulate pronouns. (22) a. b.
Si se Jeśli się if Refl "If one plays
juega mal, siempre gra zle, zawsze plays badly, always poorly, one always loses."
se się Refl
pierde. przegrywa. loses
ON IMPERSONAL REFLEXIVES IN ROMANCE AND SLAVIC
187
The impersonal is an indefinite, so it displays quantificational variability. In the conditional sentences in (22-24), the impersonal coupled with always may be equivalent to everyone, as in (22). When coupled with usually, it may be equivalent to many people, as in (23), and when coupled with seldom it may be equivalent to few people, as in (24). On this view, (22-24) are roughly equivalent in truth conditions to the relative constructions in (25-27). (23) a. b.
(24) a. b.
(25) a. b. (26) a. b. (27) a. b.
Si se juega mal, normalmente se pierde. Jesli się gra žie, zazwiczaj siq przegrywa. if Refl plays badly, usually Refl loses "If one plays poorly, one usually loses." Si se juega mal, raramente se pierde. Jesli siq gra zle, rzadko siq przegrywa. if Refl plays badly, seldom Refl loses "If one plays poorly, one seldom loses." Todo el que juega mal pierde. Wszyscy, którzy graj ą. žle przegrywają. everyone who plays poorly loses Muchos que juegan mal pierden. Wielu ludzi, którzy grają. žle przegrywają. many people who play poorly lose Pocos que jueguen mal pierden. Niewielu ludzi, którzy graj ą žle przegrywają. few people who play poorly lose
Indefinites may display variability when disclosed by adverbs of quantification, as in (22-24), but have intrinsic existential force. Disclosure is optional, so in a conditional sentence with an adverb of quantification as in (28), the impersonal may still retain existential force. When each impersonal clitic preserves its force, the disjoint reference reading arises; that is, those who explain the problems and those who do not understand need not be the same. If impersonals were free variables receiving force from the adverb, they would be semantically forced to covary in this sentence. (28) a.
Si se explica los problemas tan mal, normalmente no se los if Refl explains the problems so bad, usually not Refl them entiende. understands
MARIA LUISA RIVERO
188 b.
Jeśli tłumaczy się ćwiczenia tak žle, zazwyczaj nie if explains Refl problems so badly, usually not rozumie się ich. understands Refl them "If one explains problems so badly, usually they are not understood."
In sum, this briefest of illustrations with Spanish and Polish of some features in Chierchia's proposal highlights the semantic parallelism of the Slavic and the Romance impersonals. Combined with the syntactic similarities of section 1, the shared semantics suggest an almost complete identity between the three languages. The next section looks at a contrast that breaks this parallelism, arguing that it is due to parametric variation in the semantics of the impersonal as indefinite. 3. The impersonal and semantic variation Spanish and Polish "bare" impersonal constructions such as (3) partially repeated now as (29a) and (30a) share syntax and semantics. (29) a.
b.
(30) a.
b.
Antes se leía estos libros con placer. before Refl read.3S these books with pleasure "Before {one/people} read these books with pleasure." Se trabaja bien. Refl work.3S well "One/ people work(s) well." Tęksi^zkq czytało się zprzyjemnosci^. this book.ACC read.NEU Refl with pleasure "{One/people} read this book with pleasure." Pracuje się dobrze. work. 3 S Refl well "One/ people work(s) well."
However, when unselected Datives are added, as in (31), a clear contrast arises. This contrast also affects intransitives, as in (29b-30b) vs. (32). (31) a.
Antes se me leía estos libros con placer. before Refl me.DAT read.3S these books with pleasure "Before {one/people} read these books to me with pleasure."
ON IMPERSONAL REFLEXIVES IN ROMANCE AND SLAVIC
b.
(32) a.
b.
189
Tę książkę czytało mi się z przyjemnością. this book.ACC read.NEU me.DATRefl with pleasure Dominant reading: "I (Agent) read this book with pleasure." Dispreferred but possible: "'{One/people} read this book to me with pleasure." A Juan se le trabaja bien. John.DAT Refl him.DAT work.3S well "One works well on John's behalf." Jankowi pracuje się dobrze. John.DAT work.3S Refl well "John works well."
Spanish and Polish share a use of Datives called the Ethical Dative or the Dative of Interest. However, in Spanish (31a) and (32a), the unselected Dative has no effect on the impersonal, while in Polish it strongly suppresses the impersonal reading. In (31b) and (32b), the much preferred interpretation is with the Dative as Agent. That is, the speaker read a precise book in (31b), and John is a good worker in (32b). The Polish Dative strategy extends to sentient beings, as in (33a-b), and excludes inanimates, as in (33c). (33) a.
b.
Koniom dziś siq źle ciagnqlo wóz. horses.DAT today Refl badly pull.NEU cart.ACC "The horses had a hard time pulling the cart today." Psu się niewygodnie śpi w obro ży. dog.DAT Refl uncomfortably sleeps in collar "The dog must be uncomfortable sleeping in a collar." *Pólce ciqzko siq wisiało na cianie. shelf.DAT hard Refl hang.NEU on wall Intended: "*The shelf had difficulty hanging on the wall."
Example (33a) has a syntactic parallel in Spanish (34a). This has the (odd) reading that someone experienced difficulty when pulling the cart for the benefit of the horses; that is, Spanish impersonal se retains its flavor, and the Dative phrase is a Benefactive. The idea in (33a) could be expressed as in (34b) with "the horses" in the Dative. As in Polish (33a), in (34b) either personal feelings are attributed to the animals, or the speaker commiserates with them. However, (34b) is not a construction with (impersonal) se, highlighting the difference between Polish and Spanish in this grammar area.
MARIA LUISA RIVERO
190 (34) a.
b.
A los caballos hoy se les tiró the horses.DAT today Refl they.DAT pulled.3S mal del carro. bad of.the cart "Today someone had a hard time pulling the cart for the horses." A los caballos hoy les fue mal the horses.DAT today they.DAT went bad al tirar del carro. at.the pull of.the cart "Today, things went bad for the horses when pulling the cart."
How can such a striking contrast come about? Rivero (1999) suggests that the Polish and Romance impersonals are pronouns in the sense of the discussion in 2.1; in Polish, this pronoun has an expletive use absent from Romance. Here, I elaborate on this suggestion, proposing that it represents parametric variation in the semantics of the impersonal as indefinite pronoun in the sense of 2.2. The idea is that the Polish impersonal can be disclosed not only by adverbs of quantification, which makes it similar to its Italian and Spanish counterparts, but also by unselected Datives, which distinguishes it from Romance (and other Slavic languages (Rivero 1999)). In this analysis, the contrast in (31-32) combines the following aspects. One, it is located in the semantics. Two, it is located in the impersonal use of the reflexive clitic, which is a functional category. Three, it consists of a language specific choice in Polish of discloser-binder. Namely, in this language Datives are added to the UG list of disclosers. Let us begin by showing that the contrast in (31-32) is closely tied to the impersonal -i.e. się as syntactic and semantic indefinite pronoun--, and not just to the Dative of Interest. Recall that the anticausative or inchoative use of the reflexive in (35) is shared by Romance and Slavic, including languages without the impersonal such as Bulgarian. This use best observed in episodic sentences, displays a NOM NP and a predicate that agrees with it in PHIfeatures, and crucially lacks a {human/sentient} Agent or Causer. (35) a. b.
Se rompieron las gafas. Refl broke.PL the glasses Zlamaly się okulary. broke.FEM. PL Refl glasses.FEM.PL.
ON IMPERSONAL REFLEXIVES IN ROMANCE AND SLAVIC
c.
Sčupixa se očilata. broke.PL Refl. glasses.the 'The glasses broke."
191
Bulgarian
Spanish, Polish, and Bulgarian share a productive Ethical Dative, which can combine with this anticausative reflexive, as in (36). (36) a. b.
A Juan se le rompieron las gafas. John.DAT Refl him.DAT broke.PL the glasses Jankowi zlamaly się okulary. John.DAT broke.FEM.PL Refl glasses.FEM.PL Na Ivan mu se sčupixa ocilata. John.DAT him.DAT Refl broke.PL glasses "The glasses broke on John" or "John accidentally broke his/the glasses."
The semantic effect of the Dative in (36) is the same in the three languages. John is somehow affected by the breaking event, but need not be either the Agent or the Possessor of the glasses. Thus the Ethical Dative strategy in (36) is shared, but it differs in important ways from the Polish strategy in (31-32), which is not shared, and in my view relies on the language-specific disclosure of the impersonal. Several differences separate the constructions in (36), but they do not affect this basic conclusion. One is that clitic doubling is possible for the Dative in Spanish and Bulgarian, but does not exist in Polish, which fails to affect the semantics. Another difference is the position of the reflexive within the clitic cluster; Spanish places it first, which is not the case in Bulgarian or Polish (see (31b)). Again this has no effect on the shared meaning. Finally, each language places clitics differently in the clause. Bulgarian is a ToblerMussafia language, which the other two are not, but this has no effect on the different uses of the reflexive clitic. Bulgarian, Polish, and Spanish share constructions with psychological Vs, Datives as Experiencers, and reflexives. One example is Spanish A Juan se le olvidaron los libros "John forgot the books." This type of Dative Experiencer bears a resemblance to so-called Quirky Subjects. However, the strategy in (32-33) is again different, and one telling contrast is that when the Dative combines with the impersonal, it inherits the Thematic role that would be assigned to się if unaccompanied (Agent, or Theme). Thus, the number of Thematic roles and their nature are both preserved, which is also the case
RI LUISA RIVERO
192
when indefinites are disclosed by adverbs of quantification. I see this aspect as one consequence of disclosure in the case of (32-33). In sum, the strategy that combines an unselected Dative with the impersonal in Polish does not reduce to more familiar effects with Datives of Interest or Quirky-like subjects, which Polish shares with Spanish. The unselected Dative that in Polish combines with the impersonal clitic seems to borrow the formal role of that clitic, but not its "referential" character, and I propose to view it as a discloser in the way that follows. a) In the general case, disclosers for the impersonal in Romance and Slavic are the adverbs of quantification. They provide varying quantificational force, as discussed in 2.2. and illustrated in (22-24), but do not affect the impersonal in other ways, and Theta-roles are preserved. b) In Polish, non-selected Datives may also function as disclosers. They wipe out the intrinsic existential force of the impersonal and come to bind it, as in (31b) and (32b), without affecting Theta-roles. In this strategy, the Dative is comparable to a Left Dislocated phrase, and the reflexive is like a resumptive clitic, but is indefinite. We could thus call the Polish construction in (31-33) the Impersonal Clitic Left Dislocation. c) The interpretable content of the impersonal amounts to a {human/sentient} feature, and this is also preserved under disclosure. The Dative must match this feature in order to disclose the impersonal, which excludes inanimates from the construction. For Chierchia, the human feature in the impersonal functions as a semantic index. Thus, it can be assumed that the discloser operator on the Dative is the human feature treated as an index as well, when it matches the index of impersonal siq. A Dative that is missing this index, i.e. is inanimate, lacks the disclosure operator so cannot affect the impersonal. On this view, the semantics of Polish Dative disclosure for (32b) Jankowi pracuje siq dobrze. "John works well." is as in (37), which I owe to Gennaro Chierchia: (37) a. b.
d. e.
John.DAT [hum] work.3S siq [hum] well. siq (work) = x[hum] [work(x[hum])]. John.DAT [hum] work.3S się [hum] DIS x[hum] [hu ] [work(x[hum])] (John). Ax[hum] [work(x[hum])] (John). work (John).
ON IMPERSONAL REFLEXIVES IN ROMANCE AND SLAVIC
193
(37b) represents the semantics of the "bare" impersonal in Italian, Polish, or Spanish. The semantic feature that is represented as an index is notated [hum]. (37c) is for Polish: the human variable is disclosed via the discloser operator [hum] in the Dative, and the result is predicated of John. By the semantics of disclosure, (37c) is equivalent to line (37d). (37e) is the simplified version of (37d) after lambda conversion. Thus, the effect of the disclosure procedure is to eliminate the impersonal from the structure, while retaining other aspects. The proposed disclosure analysis makes a variety of (correct) predictions that I cannot explore for lack of space, but I will mention two in closing. The first one is that a John.DAT' + się type expression will now be able to antecede ordinary pronouns including little pro, but not another bare impersonal, which will have its own existential force. This is a correct prediction. The second prediction suggested to me by Chierchia is that there should be quantificational variability when the discloser is an indefinite. That is, a construction of type (A) linguist.DAT seldom się works well should be equivalent to Few linguists work well, and other adverbs of quantification should contribute different forces. As far as I can see, this also seems correct. 4.
Conclusions Impersonal se/si/siq is a syntactic Simplex Expression anaphor: a null expression whose defective interpretable content amounts to a human feature, and which raises to the clitic to repair formal and referential deficiencies. It repairs its formal deficiency by checking Case, and its referential deficiency by acquiring quantificational force (i.e. in R&R's terms, it forms a Casemarked chain, and goes from -Ref to +Ref), while remaining without PHIfeatures. It can function as syntactic controller, syntactic binder of all types of anaphors, including the long distance type, and reciprocals. It can be an external/internal argument of many predicates, including those associated with individual variables. The impersonal is a non-reflexivizer, so similar to a regular pronoun it can be free. Semantically, the impersonal is a human indefinite pronoun with existential force. As a pronoun, it can escape the Novelty Condition and antecede itself, which is not possible for indefinite NPs. The impersonal is existential but displays (optional) quantificational variability when adverbs of quantification disclose it and bind it. Thus the impersonal can be equivalent to some and all, as many in the past have noted, but also to many and few, which is less known. Polish offers a semantic parametric choice that affects the clitic as functional category, and is absent from Romance (and other Slavic languages). In Polish not only adverbs of
194
MARIA LUISA RIVERO
quantification but also unselected {human/sentient} Datives may disclose indefinite siq, which thus functions as an expletive pronoun that transmits properties to its discloser. On this view, the differences between the Romance and Polish impersonal arises from parametric variation in semantics. They derive from a choice of discloser for the impersonal as indefinite. They reside in the reflexive clitic as functional category, which is the type of item also considered the locus of syntactic variation.
REFERENCES Belletti, Adriana. 1982. "Morphological Passive and Pro-drop: The Impersonal Construction in Italian". Journal of Linguistic Research 2.1-34. Chierchia, Gennaro 1995. "The Variability of Impersonal Subjects". Quantification in Natural Language ed. by E. Bach, E. Jelinek, A. Kratzer, & . Hall Partee, 107-143. Dordrecht: Kluwer. Chomsky, Noam. 1981. Lectures on Government and Binding. Dordrecht:
Foris. . 1998. "Minimalist Inquiries: the Framework". To appear in Step by Step ed. by R. Martin, D. Michaels, & J. Uriagereka. Cambridge, Mass.: MIT Press. . 1999. "Derivation by Phase". MITms. Cinque, Guglielmo. 1988. "On Si Constructions and the Theory of Arb". Linguistic Inquiry 19.521-583. Dobrovie-Sorin, Carmen. 1998. "Impersonal se Constructions in Romance and the Passivization of Unergatives". Linguistic Inquiry 29.399-438. Dziwirek, Katarzyna. 1994. Polish Subjects. New York: Garland. Heim, Irene. 1982. The Semantics of Definite and Indefinite Noun Phrases. Ph.D. Dissertation. Amherst, Mass.: GLSA. Jaeggli, Osvaldo. 1986. "Passive ". Linguistic Inquiry 17.587-622. Kański, Zbigniew. 1986. Arbitrary Reference and Reflexivity: A Generative Study of the Polish Pronoun się and its English Equivalents. Uniwersytet Slaski, Katowice. . 1992. "Impersonal constructions as a Strategy for Second Order Predication". Meaning and Grammar ed. by M. Kefer & J. van der Auwera, 95-121. Berlin: Mouton de Gruyter. Mendikoetxea, Amaya. 1999. "Construcciones con se: Medias, Pasivas, e Impersonales." Gramática Descriptiva de la Lengua Española ed. by I. Bosque & V. Demonte, Vol. 2.1631-1722. Madrid: Espasa-Calpe.
ON IMPERSONAL REFLEXIVES IN ROMANCE AND SLAVIC
195
Nunes, Jairo. 1991. "Se Apassivador e Se Indeterminador: O Percurso Diacrônico no Português Brasileiro". Cadernos de Estudos Lingüísticos 20. 33-58. Perlmutter, David. 1971. Deep and Surface Structure Constraints in Syntax. New York: Holt, Rinehart & Winston. Reinhart, Tanya & Eric Reuland. 1993. "Reflexivity". Linguistic Inquiry 24. 657-720. Rivero, María Luisa. 1972."On Conditionals in Spanish". Generative Studies in Romance Languages ed. by J. Casagrande & B. Saciuk, 196-214. Rowley, Mass.: Newbury House. . 1999. "On Impersonal się in Polish: a Simplex Expression Anaphor". To appear in the Journal of Slavic Linguistics 8.1. Volume on Polish Syntax ed. by P. Bański & E. Willim. & Milena Sheppard. 1999. "On Impersonal se/siq in Slavic." To appear in Proceedings of Formal Description of Slavic Languages 3. December 1999. Leipzig, Germany. Růžička, Rudolph. 1992. "Slavic and Italian Impersonal Constructions with Reflexive Clitics". Fügungspotenzen Zum 60 Geburtstag von Manfred Bierwisch ed. by I. Zimmerman, 133-161. Berlin: Akademia Verlag. Schenker, Alexander. 1988. "Slavic Reflexive and Indo-European Middle: a Typological Study". American Contributions to the 10th International Congress of Slavicists ed. by A. Schenker, 363-383. Columbus: Slavica. Siewierska, Anna. 1988. "The Passive in Slavic". Passive and Voice ed. by M. Shibatani, 243-289. Amsterdam & Philadelphia: John Benjamins.
THE ROLE OF SYNTACTIC MODIFICATIONS ON L2 ORAL COMPREHENSION* NURIA SAGARRA The University of New Mexico
0.
Introduction The role of input in second language acquisition (SLA) has been the focus of much SLA research and theory for the last two decades. Input processing depends on the second language learners' limited capacity to process and store incoming information (McLaughlin 1987, VanPatten 1996). If the demands of a task exceed what learners can attend to at a given point in time, either processing will slow down or storage will decrease, resulting in poor comprehension. This phenomenon is especially evident at early stages of acquisition, because limited knowledge and expectations about the second language demand a greater amount of cognitive effort. One way of reducing this cognitive effort to make both input comprehensible and communication possible consists of simplifying the input (Hatch 1983). According to the Input Hypothesis (Krashen 1980, 1982, 1985), SLA needs the presence of comprehensible input to occur. Input becomes comprehensible by means of linguistic simplifications and modifications to discourse structure (e.g., negotiated interaction) (Long 1983, 1985; Pica, Young & Doughty 1987). If linguistic and interaction adjustments promote comprehension of input, and comprehensible input promotes acquisition, it could be deduced that such modifications indirectly promote acquisition (Long 1985). The aim of this study is to examine the role that simplified syntax plays in facilitating comprehension in the second language classroom with elementary learners. The present study, thus, investigates the nature of teacher talk rather than * I would like to thank Craig Chaudron and Bill VanPatten for their suggestions on earlier versions of the experimental materials. Thanks also to Garland Bills, Paola Dussias, John Lipski, Rena TorresCacoullos, the anonymous reviewers, and the audience that attended the presentation at the 30th LSRL Symposium for their comments on the paper. Any shortcomings of the study are my sole responsibility.
198
NURIA SAGARRA
foreigner talk or native talk (see Freed 1995 and Terrell 1995 for a discussion of these terms). 1. The role of morpho-syntactic modifications on oral comprehension Rubin (1994) classifies linguistic input modifications into acoustic and morpho-syntactic modifications. Acoustic modifications consist of speech rate reduction, hesitation increase, division of speech, adjustments to rhythmic patterning perception, and variation of pause duration, distribution and frequency. On the other hand, morpho-syntactic modifications consist of simpler morphology, more redundancy and discourse markers, variations in word order, and less complex syntax. Although, to the author's knowledge, there is no research investigating the effects of morphological complexity on second language comprehension, there is a body of literature on the effects of the other morpho-syntactic modifications. 1.1 Redundancy Redundancy (e.g., sentence repetition) has been found to improve comprehension at intermediate and advanced levels of proficiency (Cervantes and Gainer 1992, Chaudron 1983, Chiang & Dunkel 1992, Glisan 1985, Pica, Young & Doughty 1987). In Chiang & Dunkel's (1992) study, 180 intermediate and 180 advanced Chinese learners of English listened to a combination of -/+ modified (through redundancies and elaborations) and general knowledge of a -/+ familiar topic. Results showed that topic familiarity was important for all subjects, and that redundancies and elaborations helped comprehension in advanced learners. Glisan (1985) also found that advanced students understood longer and redundant sentences better than unmodified sentences. Pica, Young, and Doughty (1987) reported similar results when intermediate learners were exposed to sentences with repeated content words. Although the positive role of redundancy with intermediate and advanced learners is unanimously accepted, there is controversy over the effects of repetition at elementary levels of proficiency. Cervantes and Gainer (1992) found that both beginners and intermediate learners benefited from both less subordination and more repetition. However, they mixed first- and fourthsemester Japanese learners of English for the statistical analyses. Other researchers (e.g., Lynch 1988) suggest that redundancy does not aid comprehension in beginners, because learners at the lowest level of proficiency are not able to discern alternative from additional information.
SYNTACTIC MODIFICATIONS AND L2 COMPREHENSION
199
Consequently, redundant information is processed as new information and increases the listening load. 1.2 Discourse markers Chaudron and Richards (1986) differentiated two types of discourse signals: macromarkers—which indicate the overall structure of a text through emphasizing the major information and the organization of that information— and micromarkers—which act as fillers or sentence linkers. They presented 71 pre-university and 81 university Japanese students of English with one of four versions of a video tape and transcript of a lecture about U.S. history: unmodified, with micro markers, with macro markers, or with both micro and macro markers. Results revealed that the macro version helped comprehension more than any other version. 1.3 Word order According to VanPatten (1996), "second language learners are known to assign the first noun of an input sentence the role of subject even if it is something else, reflecting some universal principles of semantic role assignment" (p. 88) (for additional research, see Binkowski 1992, Gass 1989, Lee 1987, VanPatten 1984, and VanPatten & Cadierno 1992). Glisan (1985) found similar results when she investigated the listening comprehension of advanced learners of Spanish using texts with various word order patterns. Results revealed a hierarchy of comprehension: (1) subject-verb-object, (2) verb-subject-object, and (3) object-verb-subject. Thus, the aforementioned studies provide evidence that syntactic structures where the first noun of a sentence or clause functions as the subject are easier to comprehend than those where the first noun acts as the object. 1.4 Syntactic modifications As illustrated in Table 1, there is controversy over what kinds of syntactic modifications improve oral comprehension1. Cervantes & Gainer (1992) limited their study of syntactic modification to including redundancies and lowering the degree of subordination. They conducted two experiments with Japanese learners in their first and fourth semester of English. In the first experiment, subjects listened to a version with or without subordination.
1
In reading comprehension research, we also find studies in favor of syntactic simplifications (e.g., Johnson 1981; Yano, Long & Ross 1994) and against them (e.g., Blau 1982; Leow 1993; Parker & Chaudron 1987)
NURIA SAGARRA
200
Studies
Subjects
Modifications
Measure
Results
Blau (1990)
72 Polish EFL2 100 Spanish EFL Mixed levels
A. simple sentences B. complex s.3 + clues complex s. - clues
multiple choice
A=B A=C B=C
Cervantes & Gainer (1992)
76 univ.4 ESL2 Mixed levels
A. - subordination B. + subordination
cloze test
B>A
82 univ. ESL Mixed levels
A. - subordination B. + subordination repetition +
fill in the blanks
B>A OA C>B
Kelch (1985)
26 univ. ESL5 intermediate
A. -mod.6, - slow B. -mod., + slow + mod., - slow D. + mod., + slow
dictation
B>A B>C D>A D>C
Long (1985)
34 univ. ESL
A. - modified B. + modified
multiple choice
B>A
Loschky (1994)
41 univ. ESL Mixed levels
A. - mod.,- interaction B. + mod., - interaction + mod., + interaction
sentence verification
OA C>B
S agarra (1998)
279 univ. SFL
A. - modified B. slower more pauses D. simpler syntax
free recall multiple choice
D>A
Speidel et al. 60 second-graders. (1985) 30 with H-E6 30 with S-E6
A. H-E7/S-E6 complex syntax multiple B. H-E6/S-E6 simpler syntax choice . / -/+ H-E pronunciation
A=B A=C B=C
Table 1: Studies of the effect of syntactic modification on L2 oral comprehension 2
EFL=English as a Foreign Language. ESL = English as a Second Language s. = sentences. 4 univ. = university 5 SFL = Spanish as a Foreign Language. 6 mod. = modified H-E = Hawaiian-English language code. S-E = Standard English language code. 3
SYNTACTIC MODIFICATIONS AND L2 COMPREHENSION
201
In the second experiment, subjects listened to a version with subordination, with no subordination, or with both subordination and repetition. Cervantes & Gainer found that decreasing the degree of subordination and increasing repetition helped oral comprehension in both experiments. However, these results are invalidated by the sampling error of mixing subjects with different levels of second language proficiency in the statistical analyses. Adopting a different perspective, Long (1985) simplified the input by including longer words, syntactically less complex sentences, rephrase, and restatements. In his study, 34 non-native speakers of English of different linguistic backgrounds listened to a modified or unmodified version of a text. Results showed that the combination of syntactic modifications led to better comprehension. However, these results could be due to the fact that the modified version was slower and presented with a clearer articulation. Sagarra (1998) conducted a study with 279 beginning learners of Spanish to examine the effects of speed reduction, increase of pause duration and frequency, and syntactic simplifications on oral comprehension. She found that neither slowing down nor inserting more and longer pauses 8 facilitated oral comprehension, and that only simplified syntax was beneficial. In this study, syntax was simplified by means of less subordination, avoidance of OVS structure and shorter sentences, following Chaudron (1988). In contrast with the studies reviewed above in favor of implementing syntactic modifications to improve oral comprehension, there are a number of studies against the benefit of syntactic modifications (Blau 1990, Kelch 1985, Loschky 1994, Speidel et al. 1985). In two experiments, Blau (1990) found syntax not to be a determinant factor in oral comprehension. In the first experiment, three versions were divided into slow and normal speeds: version 1 contained simple sentences, version 2 complex sentences with surface clues to underlying structure, and version 3 complex sentences without optional surface clues to underlying structure. No significant results were found, except that one of the groups better understood version 1 in the slow format. In light of these results, Blau decided to separate normal from slower rate and added another factor (viz., with pauses) using similar subjects. Results showed that insertion of pauses increased comprehension. However, it is noteworthy that, because Blau did not specify the proficiency level of one of 8
The effect of pauses was determined by the type of post hoc test used. According to a Scheffé post hoc test, the difference between the unmodified version and the version with more and longer pauses was not significant (p = .063). However, according to the post hoc tests of Bonferroni and Tukey, such difference was significant (p = .042 and p = .034, respectively).
202
NURIA SAGARRA
the two groups of subjects (Puerto Ricans) and did not evaluate their second language proficiency with the same test9, the results from her two experiments cannot be generalized to learners of a specific proficiency level. Methodological limitations also restrict the validity and reliability of the findings of Speidel et al. (1985). In their study, children from both English and Hawaiian-English linguistic backgrounds listened to one version of three passages with two conditions: (a) standard English, standard English with Hawaiian pronunciation, or Hawaiian English, and (b) simple or complex syntax. Speidel et al. found that neither the use of simple syntax nor the use of Hawaiian-English pronunciation aided comprehension of standard English by Hawaiian-English children. Nonetheless, they had a very low sample size per version (n = 10), they did not distinguish between sentence length and subordinate clauses, and the materials displayed in their appendix reveal that the simplified passage contained almost half the number of shorter-sentence than subordinate-clause simplifications. The simplified passage thus had no subordinate clauses. This is important because the lack of subordinate clauses may have obscured the beneficial effect of shorter sentences. Loschky (1994) also concluded that syntactic modifications do not aid comprehension. In his experiment, 41 Japanese intermediate learners of English listened to a passage with modification and interaction, with modification and no interaction or with interaction and no modification. Modification was defined as elaborations and syntactic simplifications. Results showed that interaction and negotiation of meaning, rather than syntactic modifications, had an effect on oral comprehension. Finally, Kelch (1985) carried out an experiment with 26 intermediate learners of English who listened to a passage presented in one of four versions: unmodified, slower, syntactically simpler, or both slower and syntactically simpler. Kelch concluded that slowing down the flow of speech facilitates comprehension regardless of syntactic modifications. In conclusion, there is evidence that second language learners benefit from the use of (a) redundancy at intermediate and advanced levels of proficiency (Cervantes and Gainer 1992, Chaudron 1983, Chiang and Dunkel 1992, Glisan 1985, Pica, Young and Doughty 1987); (b) macro discourse markers (Chaudron and Richards 1986); and (c) avoidance of OVS structures at intermediate levels of proficiency (e.g., Glisan 1985, VanPatten & Cadierno 1993). However, comparison among studies is difficult because 9
Blau (1990) used the English entrance exam of a university in Poland for the Polish sample, and the English as a Second Language Achievement test (ESLAT) for the Puerto Rican sample.
SYNTACTIC MODIFICATIONS AND L2 COMPREHENSION
203
some studies mixed levels of second language proficiency (e.g., Cervantes & Gainer 1992), used non-standardized comprehension measures, included more than one modification per variable (e.g., Speidel et al. 1985), and implemented different syntactic modifications. For example, Kelch (1985) used synonymy, hypernymy, paraphrase, and parallel syntactic structures; Loschky (1994) modified syntax through locative sentence structures with semantic reversals; Blau (1990) did not specify how she converted simple into complex sentences; and Cervantes & Gainer (1992) decreased the number of subordinate clauses. 2. Research questions and hypotheses Based on these limitations, there is still a clear need for further research to determine what syntactic modifications help oral comprehension at various levels of proficiency. The present study attempts to cover part of this lacuna by addressing the questions below: Are 1. 2. 3.
there any oral input comprehension effects for L2 beginners when... ... subordination is excluded? ... shorter sentences are included? ... OVS structure is avoided?
Although the research to date does not provide clear evidence for or against the use of syntactic modifications as an aid for oral comprehension, to predict anything other than a positive effect of the aforementioned syntactic modifications would seem incongruent with Long's (1983, 1985) Input and Interaction Hypothesis. Thus, the research hypotheses are the following: Hypothesis 1. Lack of subordination facilitates oral comprehension. Hypothesis 2. Shorter sentences lead to greater oral comprehension. Hypothesis 3. Avoidance of OVS structure aids oral comprehension. 3. Method 3.1 Overview of research To investigate the preceding questions and overcome methodological limitations found in previous research, the present study incorporated a within-group counterbalanced design at the sentence level. First, a withingroup design with all subjects exposed to all versions ensures subjects' homogeneity (e.g., group A listens to sentence 1-version 1, sentence 2version 2, etc.; group listens to sentence 1-version 2, sentence 2-version 3, etc.; group listens to sentence 1-version 3, sentence 2-version 4, etc.).
NURIA SAGARRA
204
Second, a counterbalanced design with sentences presented in various orders increases reliability. Third, an experiment conducted at the sentence level rather than the discourse level leads to a more accurate control of the variables' effect and thus to more valid results. To this author's knowledge, the aforementioned three methodological improvements have not been simultaneously implemented by any published studies on modified input. 3.2 Subjects The subject pool consisted of 395 native speakers of English in their second semester of study of Spanish at the University of Illinois. Although whole classes were used during data compilation, only those subjects whose language of communication at home was exclusively English, and who had translated and listened to all target sentences were included in the study. As a result, the final sample size consisted of 382 subjects. All subjects had been instructed with the same communicative-based methodology with a focus on input. 3.3 Materials All sentences were recorded at the same rate and kept at a similar length (14-25 words), topic (familiar), and vocabulary covered in class. Practice sentences and distractors were constant for all subjects, and four versions of each target sentence were prepared: (1)
Unmodified version: + subordination, + complex, + OVS Yo salía con Juan que tenía 23 años, y a él I used-to-go-out with John who had 23 years, and to him-D.O. lo conoció tu padre en la fiesta de María. him-D.O. met your dad at the party of Mary. "I used to go out with John who was 23 years old, and your dad met him at Mary's party."
(2)
No subordination version: - subordination, + complex, + OVS Yo salía con Juan. Él tenia 23 años, y a él I used-to-go-out with John. He had 23 years, and to him-D.O. lo conoció tu padre en la fiesta de María. him-D.O. met your dad at the party of Mary. "I used to go out with John. He was 23 years old, and your dad met him at Mary's party"
SYNTACTIC MODIFICATIONS AND L2 COMPREHENSION
205
(3)
Shorter sentences version: - subordination, - complex, + OVS Yo salía con Juan. Él tenia 23 años. A él I used-to-go-out with John. He had 23 years. To him-D.O. lo conoció tu padre en la fiesta de María. him-D.O. met your dad at the party of Mary. "I used to go out with John. He was 23 years old. Your dad met him at Mary's party"
(4)
No OVS structure version: + subordination, + complex, - OVS Yo salía con Juan que tenía 23 años, y tu padre I used-to-go-out with John who had 23 years, and your dad lo conoció en la fiesta de María. him-D.O. met at the party of Mary. "I used to go out with John who was 23 years old, and your dad met him at Mary's party."
As shown in version (1), the control version did not contain any modification features, i.e., this version consisted of a complex sentence with a subordinate clause and the presence of OVS structure. Version (2) was identical to version (1), except that the subordinate clause had been converted into a main clause. Version (3) contained the same content as version (2), but included shorter sentences. Finally, version (4) was like version (1), except that the OVS structure was replaced by the SOV structure. A female native speaker of Spanish recorded four tapes with four combinations of the sentences in a sound-proof room. Subjects were randomly presented with one of the four tapes, and each tape contained 30 sentences: 6 practice sentences, 18 distractors, and 12 target sentences (three per version). Thus, in this counterbalanced within-subjects design, each subject listened to three target sentences from version (1), three from version (2), three from version (3), and three from version (4). 3.4 Procedure and scoring The 35-minute test was administered by the researcher at the beginning of one class period in the subjects' regular classrooms. Prior to beginning, subjects were told that the research study dealt with listening comprehension. Subjects were given a set of six practice sentences, and were encouraged to ask questions about the procedure. Afterwards, subjects listened to the remaining sentences (both distractors and target sentences) with no interruptions. For each sentence, subjects first listened to the sentence in
NURIA SAGARRA
206
Spanish twice and were not allowed to take notes. Second, subjects were allotted 40 seconds to write in English everything they could remember from what they had heard. Order was not important and the researcher stressed that the purpose of the task was to recall as much information as possible. The English translation of the target sentences was scored using Carrell's (1985) idea unit analysis. Two researchers independently divided each target sentence into idea units and discussed all discrepancies until there was complete agreement between them. All versions of a sentence contained the same number of idea units. An idea unit did not have to be recalled in exactly the same words, but it had to be clear to the researchers that the learner had understood that particular idea unit. 4.
Results A descriptive analysis with means and standard deviations for the idea unit scores for the four versions of the target sentences is found in Table 2. Means show that the unmodified version had the lowest mean, closely followed by the version with no subordination. This suggests that unmodified sentences had the least number of idea units recalled and, therefore, were the most difficult sentences to comprehend. The other two versions (no complex structure and no OVS structure) seem to have a similar degree of difficulty.
Version
n
M
SD
Unmodified (k=\2) No subordination (k = 12) Shorter sentences (k = 12) No OVS structure (k = 12)
382 382 382 382
2.93 3.11 3.29 3.27
2.12 2.18 2.31 2.33
Table 2: Means and standard deviations for the idea unit scores
A repeated-measures one-way ANOVA of the idea unit scores indicates a main effect for version (F (3) = 8.17, p = .000)10. Post hoc Bonferroni comparisons reveal that (a) simple syntactic sentences like (3) yield better oral comprehension than unmodified sentences like (1) (p = .000), and (b) avoidance of OVS structure as in (4) facilitates oral comprehension more 10
Mauchly's sphericity p value was lower than .05, indicating that the sphericity assumption was met for all cases, and that values of the ANOVA correspond to the unadjusted sphericity assumed test of within-subjects effects.
SYNTACTIC MODIFICATIONS AND L2 COMPREHENSION
207
than unmodified sentences like (1) (p = .001). Contrary to our predictions, no significant differences are found between the unmodified version (1) and the version (2) with no subordination (p = .135)11. In addition, there were no significant differences between the modified versions, i.e., (2) vs. (3), (3) vs. (4), and (2) vs. (4). 5.
Discussion We posed three main hypotheses as to the effect of syntactic modifications on second language oral comprehension. Hypothesis 1 (viz., no subordination facilitates oral comprehension) was not supported: there was no significant difference between the version without subordination and the unmodified version. One possible explanation could be that the version with no subordination is not simplified enough for beginning learners, given the complexity of the rest of the sentence (there is a coordinated clause). Hypothesis 2 (viz., shorter sentences help oral comprehension) was supported: there was a significant difference between the scores of the simpler syntax version and those of the unmodified version. Hypothesis 3 (viz., avoidance of OVS structure aids oral comprehension) was supported: the version with no OVS structure was significantly easier to comprehend than the unmodified version. These results support Long (1985) and Sagarra (1998)12. From the present paper, implications can be drawn in relation to the simplification of teaching materials and second-language pedagogy. Use of simple sentences and avoidance of OVS structures are easy to implement and have the potential to significantly help non-native speakers receive the comprehensible input they need to successfully acquire a second language. Such findings, therefore, contribute to the picture of what constitutes comprehensible oral input. Further research, however, needs to be conducted to examine the effects of other syntactic modifications at various levels of proficiency on both oral and written comprehension. In addition to the effects of particular syntactic simplifications, it is sine qua non to determine the level of proficiency at which the learners will need to be exposed to more complex syntax, following Krashen's (1982, 1985) i + 1 criterion. Finally, there is a need to 11 It is noteworthy that by using post hoc tests our results are more conservative than those yielded by contrasts. Contrast comparisons reveal significant differences not only between (1) and (3) (p = .000) as well as (1) and (4) (p = .000), but also between (1) and (2) (p = .023). 12 These results could also coincide with Speidel et al. (1985) if lack of subordination had hidden the beneficial effects of shorter sentences in their study.
208
NURIA SAGARRA
investigate whether or not simplified materials are a necessary or merely a facilitative factor for successful SLA.
REFERENCES Binkowki, Donna D. 1992. The Effects of Attentional Focus, Presentation Mode, and Language Experience on Second Language Learner's Sentence Processing. Unpublished doctoral thesis, University of Illinois at UrbanaChampaign. Blau, Eileen K. 1982. "The effect of syntax on readability for ESL students in Puerto Rico". TESOL Quarterly 16.517-528. . 1990. "The effects of syntax, speed and pauses on listening comprehension". TESOL Quarterly 24.746-753. Carrell, Patricia. 1985. "Facilitating ESL language by teaching text structure". TESOL Quarterly 19.727'-752. Cervantes, Raoul & Glenn Gainer. 1992. "The effects of syntactic simplification and repetition on listening comprehension". TESOL Quarterly 26.76-770. Chaudron, Craig. 1983. "Simplification of input: Topic and reinstatements and their effects on L2 learners' recognition and recall". TESOL Quarterly 17.437-458. 1988. Second Language Classrooms: Research on Teaching and Learning. Cambridge: Cambridge University Press. & Jack C. Richards. 1986. "The effect of discourse markers on the comprehension of lectures". Applied Linguistics 7.113-127. Chiang, Chung S. & Patricia Dunkel. 1992. "The effect of speech modification, prior knowledge, and listening proficiency on EFL lecture learning". TESOL Quarterly 26.345-374. Freed, Barbara F. 1995. "Foreigner talk, baby talk, native talk". Studies in Language Learning and Spanish Linguistics ed. by Peggy Hashemipour, Ricardo Maldonado and Margaret van Naerssen, 233-245. New York: McGraw-Hill. Gass, Susan M. 1989. "How do learners resolve linguistic conflicts?" . Linguistic Perspectives on Second Language Acquisition ed. by Susan M. Gass and Jacquelyn Schachter, 183-199. Cambridge: Cambridge University Press.
SYNTACTIC MODIFICATIONS AND L2 COMPREHENSION
209
Glisan, Eileen W. 1985. "The effect of word order on listening comprehension and pattern retention: An experiment in Spanish as a foreign language". Language Learning 35.443-472. Hatch, Evelyn. 1983. "Simplified input and second language acquisition". Pidginization and Creolization of Language Acquisition, 64-86. Rowley, Mass.: Newbure House. Johnson, Patricia. 1981. "Effects on reading comprehension of language complexity and cultural background". TESOL Quarterly 15.169-181. Kelch, Ken. 1985. "Modified input as an aid to comprehension". Studies in Second Language Acquisition 7.81-89. Krashen, Stephen D. 1980. "The input hypothesis". Current Issues in Bilingual Education ed. by J. Alatis, 144-158. Washington, DC: Georgetown University Press. . 1982. Principles and Practice in Second Language Acquisition. New York: Pergamon Press. . 1985. The Input Hypothesis: Issues and Implications. London: Longman. Lee, James F. 1987. "Morphological factors influencing pronominal reference assignment by learners of Spanish". Language and Language Use: Studies in Spanish ed. by T. A. Morgan, James F. Lee and Bill VanPatten, 221-232. Lanham, Md.: University Press of America. Leow, Ronald P. 1993. "To simplify or not to simplify". Studies in Second Language Acquisition 15.333-355. Long, Michael. 1983. "Linguistic and conversational adjustments to nonnative speakers". Studies in Second Language Acquisition 5.177-193. .1985. "Input and second language acquisition theory". Input in second language acquisition ed. by Susan Gass, 377-393. Rowley, Mass.: Newbury House. Loschky, Lester. 1994. "Comprehensible input and second language acquisition". Studies in Second Language Acquisition 16.303-323. Lynch, A. 1988. Grading Foreign Language Listening Comprehension Materials: The Use of Naturally Modified Interaction. Ph. D. Dissertation, University of Edinburgh. McLaughlin, Barry. 1987. Theories of Second Language Learning. London: Edward Arnold. Parker, Kate & Craig Chaudron. 1987. "The effects of linguistic simplifications and elaborative modifications on L2 comprehension". University of Hawaii Working Papers in ESL 6.107-133.
210
NURIA SAGARRA
Pica, Teresa, Richard Young & Catherine Doughty. 1987. "The impact of interaction on comprehension". TESOL Quarterly 21.737-758. Rubin, Joan. 1994. "A review of second language listening comprehension research". The Modern Language Journal 78.199-221. Sagarra, Nuria. 1998. "The effect of rate-controlled speech and simplified syntax on listening comprehension in beginners". Antología del Noveno Encuentro Nacional de Profesores de Lenguas Extranjeras, México D.F., México: Universidad Nacional Autónoma de México, 1998, pp. 313-322. Speidel, Gisela E., Roland G. Tharp & Linda Kobayashi. 1985. "Is there a comprehension problem for children who speak nonstandard English? A study of children with Hawaiian-English backgrounds". Applied Psycholinguistics 6.83-96. Terrell, Tracy D. 1995. "Foreigner talk as comprehensible input". Studies in Language Learning and Spanish Linguistics ed. by Peggy Hashemipour, Ricardo Maldonado and Margaret van Naerssen, 233-245. New York: McGraw-Hill. VanPatten, Bill. 1984. "Learners' comprehension of clitic pronouns: More evidence for a word order strategy". Hispanic Linguistics, 1, 57-67. . 1996. Input Processing and Grammar Instruction: Theory and Research. Norwood, NJ: Ablex. — & Teresa Cadierno. 1993. "Explicit instruction and input processing". Studies in Second Language Acquisition 15.225-244. Yano, Yakusata, Michael H. Long & Steven Ross. 1994. "The effects of simplified and elaborated texts on foreign language reading comprehension". Language Learning 44.189-219.
SYNTACTIC CONSTRAINTS ON ACCESS TO LEXICAL STRUCTURE THE CASE OF FICAR* CRISTINA SCHMITT Michigan State University
0.
Introduction There has been very little systematic work on semantic alternations in verbs that are like 'be' in taking predicates as complements and in being unable to license referential thematic arguments on their own, but semantically differing from 'be' in carrying event-structure (aspectual) information. This paper discusses the 'be'-like verb ficar in Brazilian Portuguese, which seems ambiguous between a STAY reading and a BECOME reading, as illustrated below (In the text, capital letters are used for the basic meanings). (I)
a.
b.
A familia ficou no Brasil the family ficar-PAST in Brazil "The family remained/stayed in Brazil." A cidade ficou maior. the city ficar-PAST bigger "The city became bigger."
In (la) the verb is translated as 'stay/remain' and in (lb) the verb is translated as 'become'. 1 Assuming two basic meanings STAY/REMAIN and BECOME, the standard treatment for dealing with this apparent case of lexical * Support for this work comes from a guest researcher grant from the Deutsche Forschunggemeinschaft to ZAS-Berlin. I thank Gehrard Jäger, Ewald Lang, Christine Maassen, Claudia Maienborn, Asunción Martinez-Arbelaiz, Renate Steinitz and Chris Wilder and especially Alan Munn for many important comments and suggestions. 1 Ficar also has a reading that can be translated as 'keep'. This reading arises when the complement is a gerundive form. I will not discuss this reading in this paper.
212
CRISTINA SCHMITT
polysemy is by recording each particular grammatical usage of a verb in a separate lexical entry. The relation between STAY/REMAIN and BECOME can be
encoded via meaning postulates that exploit the fact that BECOME and REMAIN are taken to be duals, i.e., REMAIN P is equivalent to NOT BECOME NOT P (e.g., Hovav & Levin 1998, Steinitz 1999). However, to create two (or more) lexical entries associated with these two basic functions for ficar not only misses speakers' intuitions that ficar is not ambiguous at all, but also misses the fact that most of what is necessary to account for the behavior of ficar in Brazilian Portuguese is given by the type of complement it takes. If the complement is a locative (adverbial or prepositional), only a STAY reading is possible. With property denoting predicates, BECOME readings are always available and STAY readings are apparently possible depending on the context, as illustrated in (2) and (3): (2)
(3)
a. b. a. b.
A Maria ficou com fame. the Maria FICAR-PAST with hunger "Maria became hungry." "Maria remained hungry." Nao era nenhuma surpresa, depots de todo aquele exercício. "It was no wonder she was hungry after all that exercise." Isso era surpreendente, depois de tudo que ela tinha comido. "This was surprising, after all she had eaten."
The sentence in (2) can be roughly translated as (2a) or (2b). Both continuations, (3a) or (3b), are felicitous and compatible with the meaning in (2a) and (2b) respectively.2 In this paper I argue that in all cases we have one verb, which basically denotes a transition, with one general argument structure and one event structure. The different readings will be explained by the interaction between the transition component of ficar and its complements. I will show the translations are partially misleading because what changes is the properties of the complement and therefore the possible interactions with ficar, whose meaning remains constant. This is in the spirit of much of Pustejovsky's (1995) generative lexicon proposal. The idea of the generative lexicon is that rather than positing a new entry for a verb every time a new environment/sense pair is discovered, one 2 Adjectives in general also behave as in (2). I will call this case the case of ficar+adjective for lack of a better term for this complex that involves APs and non-locative and nondirectional PPs and most adjectival participials.
SYNTACTIC CONSTRAINTS ON LEXICAL ACCESS
213
should "spread the semantic load" more evenly throughout the elements in composition. Such an approach partially shifts the burden of the explanation for cases like ficar+locatives from the verb itself to the locative. Moreover, a generative approach will allow us to give a uniform treatment to verbs that display the same behavior cross-linguistically, as is the case of quedar(se) in Spanish and bli in Swedish, illustrated in (4) and (5) respectively, and also to explain why verbs that have a BECOME component do not always allow STAY readings. A non-generative system of multiple entries can describe the facts but is unable to explain them. (4)
a.
b.
(5)
a.
b.
María se quedó en casa. Maria SELF quedar-PAST at home "Maria stayed at home." María se quedó sorda. Maria SELF QUEDAR-PAST deaf "Maria became deaf." Han blev i sängen Han bli-PAST in bed "Han stayed in bed." Han blev sjuk Han bli-PAST sick "Han became sick."
quedar(se) Spanish
bli - Swedish
(Steinitz 1999)
Following Jackendoff (1996), I assume that BECOME and STAY are not primitives but are rather derived from BE. The intuition I would like to pursue is that the ficar complex means the state of BE x after some change Y. Fz'car+complement denotes then a complex event composed of two subevents where Y precedes X.3 Schematically we would have the following:
Y is the subevent that encodes the transition to subevent X, in which case we get a BECOME reading. If Y cannot be interpreted as the transition to X (for reasons that will become clear below), subevent Y remains just a 3
It should be noted here that Y is not CAUSE. CAUSE predicates are associated with 'causer' arguments, but Y here is not associated with a 'causer ' argument (see Hovav and Levin 1998). It can only be associated with the argument that undergoes the change.
214
CRISTINA SCHMITT
necessary or a possible pre-condition and subevent X is all that is asserted. In this case the translation is 'remain' or 'stay'.4 The intuition I am trying to capture is exemplified in (7). Compare (7a) and (7b), in view of the following context: politician A is 'modern' and has the brilliant idea of removing the university from the downtown area so that a is not convention center and various hotels can be built. Politician 'modern' and is opposed to this idea. He can say (7a) but not (7b). (7)
a.
b.
A universidade fica aqui. the university stays here "The university will stay here." A universidade é aqui. the university is here "The university is here."
(7a) is an acceptable statement given the context but (7b) is not. (7b) just asserts that the university is here, which given the context is basically a nonsequitur, since both politicians know where the university is. (7a) illustrates what I consider to be part of the meaning of STAY, namely that for its use to be felicitous, the situation minimally could have been different. In other words, a change in the described situation was a possibility. The analysis I propose tries to account for the two basic uses of ficar which constitute the core of the empirical coverage of this paper. In more general terms, I make an attempt to define under what conditions subevents can remain unrealized. I will argue that the constraints on realization of subevents will be related to the anchoring of subevents on directly or indirectly realized arguments. In other words, subevent Y in (7) must remain unrealized because there are no arguments whose roles can be associated with it. In section 1 I present the assumptions about the lexical representations I am adopting; section 2 makes a proposal for ficar. In section 3 I discuss the behavior of ficar+adjectival predicates and in section 4 I discuss how locatives disallow BECOME readings.
4
It is beyond the scope of this paper to deal with the relation between event structure and information structure, but the idea here is that unlicensed subevents are not asserted but remain as part of the meaning of the sentence as implied and possibly presupposed.
SYNTACTIC CONSTRAINTS ON LEXICAL ACCESS
215
1. Basic assumptions Pustejovsky (1995) argues that words are used creatively and word senses are permeable in that they are not atomic definitions but overlap and make reference to other senses. If lexical representations are to account for these properties there must be a set of generative devices that connect the different levels of lexical semantics providing for the compositional interpretation of words in context. There are, according to Pustejovsky, four independent levels of representations in the lexical structure. Besides generative mechanisms that include co-composition and coercion, and a level of lexical inheritance structure, there is a level of Argument Structure, a level of Event Structure and a level of Qualia Structure. The level of argument structure specifies the number and type of arguments that a lexical item carries. This level is independent from the level of Event Structure which characterizes not only the basic event type of a lexical item but also internal subeventual structure. Finally there is a level that he calls the Qualia Structure, which represents the different modes of predication possible with a lexical item. The function of the Qualia structure level is to unpack some of the idiosyncratic/constant meaning of a word, and associate bits of that information with the argument structure and the event structure. This will allow these bits of information to compose with others in the syntax. Event Structures are at most binary and composed of sorted subevents: processes, transitions or states. The subevents can be combined by one of the following relations: precedence (<), overlap (o) and precedence and overlap (
Subevent Licensing Condition A subevent must be linked to an argument to be syntactically realized. If not associated to a referential argument (directly or indirectly), it can not be activated in the syntax.
216
CRISTINA SCHMITT
(9)
Referential Argument Sharing Condition Two subevents only share an argument ZP if the properties of ZP selected by the predicate e2 include the properties selected by el, and el does not take a referential argument. If el and e2 are independent heads, e2 must be the complement of el.
2. The proposal The goal of this section is to define the basic components that a lexical representation of a verb, such as ficar must have and that can be easily extended to its counterparts in Spanish and Swedish. The underlying idea here is that ficar is like 'be': its role is to create verbal complexes in the syntax out of predicates that cannot carry verbal tense/person morphology (PPs, APs, AdvPs, Participial Phrases and very marginally modified NPs). Also like 'be', ficar is unable to assign a thematic role to a referential argument. However, unlike 'be' which is transparent, ficar contributes with some subevent structure type which combines with the properties of the complement to derive a complete event structure. My proposal is that ficar contributes the first subevent (el) information in a structure like (7) but does not select for referential arguments (rather it takes a predicate as its complement). Ficar then combines with a predicate that contributes the (e2) information. The predicate not only has subevent properties but crucially can license a referential thematic argument. Depending on the complement, co-composition is possible and the sharing of the referential argument by el and e2 will be possible. The proposal is then the following: ficar encodes subevent information but is unable to license a thematic referential argument on its own. Since ficar is not a causative verb, I will assume that it is of category V rather than v and moreover the second subevent has to be a state (e2=state). Given this proposal the first question to be addressed is what the subevent information encoded by ficar is in the context of both ficar+locatives and ficar+adjectival predicates. I will argue that ficar denotes in both cases a transition with no duration and that the relation between el and e2 is one of Precedence. Schematically the proposal is stated in (10):
SYNTACTIC CONSTRAINTS ON LEXICAL ACCESS
217
When ficar combines with locatives, it cannot use the argument of the predicate to license its subevent, and the result is a stative reading (the transition becomes inactive in the syntax and e2 is always a state). When ficar combines with adjectival predicates (always eventive ones), then ficar can use the argument of the predicate to license its own subevent structure and a BECOME reading is available, due to the fact that el is a transition. 3. Evidence for ficar as a punctual transition First I will present two arguments to demonstrate that ficar+adjectives denote a punctual transition: the behavior of ficar+predicate when modified by time-span adverbials and the behavior of ficar+predicate when modified by the adverbial quase 'almost'. Then I will show that ficar+adjectives always denote a transition, by showing that the progressive always triggers BECOME readings.
3.1 Time-span adverbials Time-span adverbials as for x time an in x time are good tests for durative and non-durative predicates respectively (Dowty 1979; Verkuyl 1993 and references there). Consider (11) and (12): (11) a.
b. (12) a.
b.
A Maria ficou em Berlin por três dias. the Maria stayed in Berlin for three days "Maria stayed in Berlin for three days." #A Maria ficou em Berlin em dois dias. the Maria stayed in Berlin in two days A planta ficou bonita por três semanas. the plant stayed pretty for three weeks 'The plant stayed pretty for three weeks." A planta ficou bonita em urn dia. the plant became pretty in a day (The new fertilizer is very good.) "It took a day for the plant to become pretty."
We can see by the contrast between (11a) and (11b) that ficar+ locatives behaves like a durative predicate. Ficar+adjectival predicates, on the other hand, behave like an achievement verb. In (12a) we cannot have a reading in which the prettiness of the plant increases in the interval that corresponds to three weeks. Rather, the adverbial measures the result state. In (12b) the adverbial is not modifying the change of state, but rather the period that
218
CRISTINA SCHMITT
precedes the change of state, as the gloss clearly illustrates. Ficar in (12b) displays the same behavior as achievement verbs do when modified by in-xtime adverbials, as shown in (13), where 'in five minutes' modifies the preceding time of the actual reaching the summit, not the actual reaching of the summit. (13) Maria reached the summit in five minutes. "It took Maria five minutes to reach the summit." 3.2 Modification by almost Further evidence supporting the generalization that ficar+ adjectival predicates behave like an achievement verb in having a transition component with no duration is given by the behavior of quase 'almost'. When a verbal predicate like 'open the door' is modified by 'almost' we can derive three readings, as illustrated in (14). (14) a. b.
Mary almost "What Mary "What Mary "What Mary
opened the door. almost did was open the door." did was almost open the door." did to the door was almost open it."
We can assume that in (14a) we are modifying the preparatory phase; in (14b) the process; and in (14c) the result state. In the case of ficar + adjectival predicates, there are two positions in which quase 'almost' can appear: before the verb ficar or before the predicate. Here I will only consider the first position since, in the second position, quase 'almost' can only modify the predicate. When it precedes the verb, it can either take scope over the adjectival predicate (15a) or the period preceding the transition (15b). (15) a. b.
Maria quase ficou bonita/'em pânico / mais irritada. Maria almost became pretty/ in panic/ more irritated "Maria become partially pretty..." "What almost happened to Maria was to become pretty..."
If the adjectival predicate cannot be modified by 'almost', as illustrated in (16a), then only the period that precedes the event can be modified (16b). Thus in (17), the only reading we obtain is one in which Pedro did not stay at home (i.e., modification of the preparatory phase).
SYNTACTIC CONSTRAINTS ON LEXICAL ACCESS
(16) a. b.
(17)
219
A Maria quase ficou grávida. the Maria almost became pregnant #Maria almost became pregnant and the result is half pregnancy. "What almost happened to Maria was to become/stay pregnant."
Pedro quase ficou em casa. the Pedro almost stayed at home "What almost John did was to stay at home."
The behavior of ficar+predicate with quase 'almost' indicates that ficar cannot denote a gradual change, rather it denotes a punctual transition. 3.3 Ficar+adjectives and the Progressive In the introduction, I said that ficar+adjectival predicates allowed two readings: a STAY reading and a BECOME reading, depending on the context. The question is whether ficar+adj ectival predicates is really ambiguous between two readings. The answer I will give is that the ambiguity is only apparent and depends on whether we take the transition to be part of new information in the discourse or not. The argument comes from the behavior of ficar+-adjectival predicates and ficar+locatives, when selected by the progressive. In Brazilian Portuguese, unlike in English, both stative and non Stative verbs are acceptable in the progressive. (18) Maria está sabendo do problema. Maria is knowing about-the problem "Maria is acquainted with the problem." When used with an achievement, the progressive (when possible), modifies the preparation time for the event described, both in Portuguese and in English. Thus in Maria is reaching the summit, the progressive modifies the climbing event, not the reaching event. The examples in (19) show that both ficar+locative and Ficar+adjective are acceptable in the progressive.
220
CRISTINA SCHMITT
(19) a.
b.
Maria está ficando em casa. Maria is staying at home. "Maria is staying at home." Maria está ficando (mais) bonita. Maria is becoming more pretty "Maria is becoming (more) pretty."
However, the readings in (19a) and (19b) are not identical. (19a) does not have a reading in which the event is partially completed, but rather has a different reading that mainly contrasts with the present tense reading in giving a sense of temporariness to the described state. Other verbs that do not have a motion component and take locatives as arguments allow the same kind of reading, as illustrated in (20) for Portuguese and English. (20) a. b. d.
Maria está morando em Paris. "Maria is living in Paris." Maria mora em Paris. "Maria lives in Paris."
In (19b), however, only the BECOME reading is acceptable. In other words, Maria is coming closer to the state in which we can say she became pretty. Independent of the way we deal with these cases, we can assume that while there is some change within the participants of the event in (19b), there is no change in (19a). What is important here is that with adjectives, we only get a BECOME reading even when the context forces a STAY reading, as the example below illustrates, assuming a context in which Maria was very hungry at 9 o'clock but she didn't eat anything and continued to be hungry. (21) a b.
Ele estova com fome e ficou com fome. she was with hunger and stayed/became with hunger "She was hungry and stayed hungry." #"She was hungry and she became hungry."
In this context we can say (21) with the meaning in (21a). A BECOME reading (21b) would be infelicitous in this context. Now consider the same sentence in the progressive:
SYNTACTIC CONSTRAINTS ON LEXICAL ACCESS
221
(22) #Ela estava com fame e estova ficando com fame. she was with hunger and was becoming with hunger "She was hungry and was becoming hungry." Although the context forces the STAY reading of ficar, the progressive still forces the BECOME reading and wins over context. This is not predicted if ficar+adjectival predicates could freely display the BECOME or STAY meanings. I conclude from (22) that ficar+adjectival predicates always have a BECOME reading.
In order to assume that the transition portion of the complex of ficar+adjectival predicates must come from ficar, we need to rule out the possibility that the transition comes from the adjectival predicate which is always stage-level (e.g. Schmitt 1991). This can be easily done by examining the behavior of the same predicates with the copula verb estar, which also takes stage-level predicates. As shown below, the complex of estar+predicate never asserts a transition to the state described by the adjective. (23) Maria está com fame/ bonita/maior. Maria is with hunger/ pretty/ bigger "Maria is hungry/pretty/ bigger." The fact that ficar can only combine with stage-level predicates can be explained if ficar can only combine with an eventive predicate. The BECOME readings are possible because the transition el can share an argument with e2 since the Referential Argument Sharing Condition (9) can be easily satisfied. As is well-known, one of the properties of the so-called stage-level predicates is to denote a non-inherent property. If it is not inherent to the subject, the subject must have been able to undergo a transition at some point in time from not having property P to having property P, and this must be part of the properties of the subject selected by the stage-level predicate. Ficar can denote exactly the transition that precedes the state of having Property P and both can be predicated of the same argument. Assuming that ficar selects no referential thematic argument on its own, then it can easily enter co-composition with the stage-level predicate that is its complement and BECOME readings will be obligatory for the complex.5 The STAY reading is actually a discourse effect, namely it is the result of taking the transition as known in the discourse and therefore not relevant. 5 Cases where it appears that ficar is used intransitively have actually an implicit locative as the main predicate (see Schmitt 1999 for evidence).
222
CRISTINA SCHMITT
Evidence for this conclusion comes also from the fact that in cases where the context allows both readings, the so-called STAY reading is only possible if the predicate is destressed. If we assign main focus stress to the predicate only a BECOME reading is available, since the predicate is signaled as part of the new information in the discourse (see Schmitt 2000 for details). 4. Ficar+locatives: where is the transition? Clearly the same discourse explanation cannot work for ficar+locatives. If fiicar always denotes a transition and the STAY readings were in this case also discourse dependent, then BECOME readings should be possible depending on the context, but they are not. If ficar is basically a head that means TRANSITION that selects for a state/location, then why does the combination of CHANGE + LOCATIVE not give us a reading like go, get to, or arrive at, and if we never get a change of state, what happens to this transition part? 4.1 Why transitions cannot enter co-composition with locatives When we combine a transition (which can be defined as going from a negative state to a positive state) with a state, the result should be a transition to a positive state. Combining a transition with a locative should produce a transition from not being at a location to being at a location. Let's assume that locative prepositions are two-place relations. In John in the park the relation is between two objects: John and the park. We can assume that ficar merges with a PP that has a specifier and a complement. I am ignoring functional projections for this discussion. We know that ficar in these cases does not allow modification by in x time and has no BECOME readings (not even with the progressive). Therefore we can assume safely that only e2 is available for modification in the syntax. What is necessary to know is what happens to el. In the BECOME readings it is clear that el is associated with change and that the change involves a change in a property that the subject comes to have in the result state. Assuming the Referential Argument Sharing Condition (9), the question is why we cannot establish that the transition sub event is related to the external argument of the locative predicate. If we use Jackendoff s (1996) decomposition of STAY and GO, we certainly have a common basis. Both involve a relation BE between [Thing] and [Space]. The question is why ficar+locative cannot mean GO. The difference between STAY and GO, according to Jackendoff, is that in GO the location of the object changes with time and with the event but not
SYNTACTIC CONSTRAINTS ON LEXICAL ACCESS
223
with STAY. For GO, the spatial coordinates of the [Thing] change with time. In fact, the binding of the change of location by the [Thing] with [Time] creates a PATH. STAY, on the other hand, does not allow a homomorphism between the event, time and the change of location, in such a way that a PATH can be created to measure the event. We have seen already that ficar denotes a transition that has basically no duration and precedes e2. Thus, ficar is a type of change that does not allow a PATH to be created (i.e., intermediate states between not having P and having P), which would be allowed if the relation was precedence and overlap and if ficar denoted a process. On the other hand, the preposition by itself cannot create a PATH because the preposition has no directional properties. It is a locative preposition. In fact, directional prepositions, which allow a change of location with GO verbs (24a), are impossible with ficar (24b). (24) a. b.
Elefoi de Porto Alegre para São Paulo. "He went from Porto Alegre to São Paulo." *Ele ficou de Porto Alegre para São Paulo. "He stayed from Porto Alegre to São Paulo."
Even if we try to have a change that is not durational but is arguably punctual, like buy and sell in the possessional domain, the result is unacceptable, as in (25a). In this case we need a GO verb as in (25b). (25) a.
*O livro ficou da Maria para a Cláudia. the book came-to-be from-the Maria to the Cláudia. #"Maria's book is now Cláudia's book." b. * livro foi da Maria para a Cláudia. the book went from-the Maria to the Cláudia. #"Maria's book is now Cláudia's book."
Directional prepositions have a motion component encoded in their Qualia that needs to combine with a verbal element that also has a motion component in order for the result to be well formed. Moreover, directional prepositions are complex events. Assuming with Pustejovsky that Events are at most binary, we will never find verbs like ficar taking directional prepositions as complements. The conclusion is that although we have 'change' as part of the meaning of the verb ficar, the locative preposition does not select for this property on
224
CRISTINA SCHMITT
its arguments. Therefore it is impossible for the transition (el) to be licensed, assuming that in order for a subevent to be realized it must license (directly or co-compositionally) a referential argument. Since it cannot share an argument with the locative preposition, it cannot be licensed in the syntax. But now we are left with an unrealized subevent (el). In Pustejovsky's analysis of build, for example, build has three arguments, a builder, a product and also material, since we cannot build without material. Therefore 'material' is part of the logical argument structure of the verb. The material argument can be realized in various ways as illustrated below: (26) a. b.
John built a house out of wood. John built a wooden house. John built a house.
It can be realized as an adjunct (26c), or as a modifier of the product itself (26b), since the material is a CONSTITUTIVE role of the product of building. Alternatively it can be realized as an implicit CONSTITUTIVE role of 'house' in (26c) (see Pustejovsky 1995 for details). Given the Subevent Licensing Condition (8), subevent (el) needs to be associated with a referential argument in order to be licensed. The only candidate for this role is the external argument of the preposition. After all, in Maria stayed home, Maria is an individual that has as part of its Qualia the property of being movable, changeable. If we could use the movable properties of the locative preposition external argument, then we should be able to associate it to the transition, but clearly we cannot, otherwise we would get a change of location reading. Here is where Condition (9) comes to play its role. The movable properties of Maria are not the properties that are being selected by the locative head. Therefore they are invisible to ficar. In other words, the qualia of ficar cannot have access to the properties of Mary that are not a subset of the properties selected by the preposition. For this to occur the Qualia structure of the preposition should have also selected for the movable properties of the predicate. This could be the case if ficar could take directional prepositions, something that is independently ruled out by the binary condition on Complete Event Structures. We are left with the disturbing fact that we have a subevent not linked to an individual argument at all. If it cannot be realized, it consequently cannot be part of what is asserted by the proposition. It remains as part of the unpacked meaning of ficar, as illustrated in (7).
SYNTACTIC CONSTRAINTS ON LEXICAL ACCESS
225
Of course we need to constrain the conditions under which subevents can remain unrealized. I would like to speculate that the only kinds of subevents that can be left unlinked to an argument are transitions that must be interpreted as punctual. This makes the prediction that BECOME verbs that allow gradual change, do not display STAY readings with locatives and BECOME verbs that allow gradual change display STAY readings. The prediction is borne out for quedar(se) (Spanish), bli (Swedish), werden (German) and get (English). While the former two disallow gradual transition readings, German and English allow gradual change readings. While the Spanish and the Swedish counterparts of ficar display BECOME and STAY readings (with adjectival predicates and locatives, respectively), the English and the German verbs disallow locatives and STAY readings altogether. Since the change part allows gradation in time in English and German, the sub event cannot be left unrealized and the result is ungrammatical (see Schmitt 2000).
REFERENCES Dowty, David. 1979. Word Meaning and Montague Grammar. Dordrecht: Kluwer. Hovav, Malka Rappaport & Beth Levin. 1998 "Building verb meanings." The Projection of Arguments: lexical and compositional factors ed. by Miriam Butt and Wilhelm Geuder, 97-134. Stanford: CLSI. Jackendoff, Ray. 1996. "The proper treatment of measuring out, telicity and perhaps event quantification in English". Natural Language and Linguistic Theory 14.305-354. Pustejovsky, James. 1995. The Generative Lexicon. Cambridge, Mass.: MIT Press. Schmitt, Cristina. 1991. "Ser and estar: a matter of aspect". Proceedings of the Northeastern Linguistic Society 22.411-425. . 1999. "Ficar: bleiben and werden in one: licensing subevents". ZAS Papers in Linguistics, 14.227-255. . 2000. "Transparent and non-transparent verbalizers". Ms. Michigan State University. Steinitz, Renate. 1999. "Deutsch werden, bleiben: Swedisch bli, förbli". ZAS Papers in Linguistics 14.209-226. Verkuyl, Henk. 1993. A theory of aspectuality. Cambridge: Cambridge University Press.
A COMPARATIVE SEMANTICS FOR THE SUBJUNCTIVE MOOD IN SPANISH* ELISABETH VILLALTA University of Massachusetts at Amherst
0.
Introduction The distribution of indicative and subjunctive verb forms in Romance languages is known to be complex and rather difficult to characterize. In this paper, I attempt to provide a semantic characterization of the licensing contexts for subjunctive verb forms in Spanish. While I mainly concentrate on their distribution in the complement clause of propositional attitude predicates, ultimately, my goal is to determine an account that unifies as much as possible all the different uses of the subjunctive mood in Spanish. The proposal presented here departs from the widespread idea that a criterion related to the realis/irrealis notion plays a relevant role in the distribution of subjunctive mood. Instead, I investigate the following hypothesis: in Spanish, subjunctive mood appears in contexts that require comparison of alternative propositions. Whether these alternatives are real or counterfactual is irrelevant for the distribution of subjunctive mood. The structure of this paper is as follows. Section 1 presents the core facts about the distribution of subjunctive mood in Spanish embedded clauses and discusses briefly the line of investigation adopted in recent semantic approaches. In section 2, I examine the semantic properties of the relevant predicates. I propose to extend Heim's (1992) conditional semantics to all predicates selecting the subjunctive mood in Spanish. In section 3,I argue for a revision of
* I wish to thank Angelika Kratzer, Barbara Partee and Yael Sharvit for extensive discussion and very helpful comments. I am grateful to the participants at the 30th Linguistic Symposium on Romance Languages for useful comments.
228
ELISABETH VILLALTA
this semantics that crucially involves comparison of alternatives. Section 4 presents empirical evidence for the proposal. In section 5, extensions of the proposal to other environments are considered. 1. Recent approaches to the Subjunctive Mood in Spanish Embedded Clauses In Spanish, the indicative mood is selected by epistemic predicates {know, think, believe), declarative predicates {say, write, declare), predicates of certainty {be sure, be convinced), commissives {promise), and fiction verbs {dream, imagine). The subjunctive mood is selected by desire predicates {want, wish, hope, fear), directives {order, advise, suggest), modals {it is possible, it is likely), predicates expressing doubt {not believe, doubt) and emotive /active predicates {regret, be glad, be sorry,).1 Some examples for both classes are given in (1) through (4). (1)
(2)
(3)
(4)
Violeta cree que la fiesta comienza a las siete. Violeta believes that the party starts (Ind.) at the seven. "Violeta believes that the party starts at seven." Josefa dice que traerá la música. Josefa says that bring (Fut. Ind.) the music. "Josefa says that she will bring the music." Violeta quiere que Josefa cante. Violeta wants that Josefa sing (Subj.). "Violeta wants Josefa to sing." Violeta se lamenta de que Josefa no quiera cantar. Violeta SE regrets of that Josefa not wants (Subj.) sing. "Violeta regrets that Josefa doesn't want to sing."
Throughout the linguistic literature on Romance languages, numerous proposals for a characterization of these two predicate classes have been defended.2 The most prominent traditional approach claims that the notion of realis/irrealis plays a determining role for the distribution of subjunctive and indicative verb forms. While the indicative (the realis mood) would require the
1 I have adopted the predicate classification following Farkas (1992). 2 For an overview of recent approaches to the semantics of mood see Portner (1999). For a summary of proposals from traditional and transformational grammars see Bell (1980), Bergen (1978), Klein (1974), and Palmer (1986) for a crosslinguistic survey on mood and modality.
COMPARATIVE SEMANTICS FOR SUBJUNCTIVE MOOD
229
embedded proposition to be true in the actual world, this would not be the case for the subjunctive mood (the Irrealis mood). Since Hintikka (1969), however, there is a common understanding that all propositional attitude predicates require evaluating the embedded proposition in worlds that are not the actual world. Recent approaches (cf. Farkas 1985, 1992, Giannakidou 1998, Quer 1998, Giorgi & Pianesi 1997 and Portner 1997) have proposed a more fine-grained characterization of the contexts of evaluation by using the tools provided by recent developments in possible world semantics. I will briefly present their main insights in what follows. Farkas (1985,1992) proposes a classification into weak and strong intensional predicates, by introducing the notion of truth of a proposition relativized to individuals. She proposes that propositional attitude predicates select the indicative mood if the proposition is true in the real world as far as the subject is concerned (i.e., if the subject believes in the actual world that the embedded proposition is true). Subjunctive mood is selected otherwise. Similarly, Giannakidou (1998), building on Farkas' insight, proposes to subsume the relevant criterion under the notion of context-dependent veridicality (also adopted in Quer 1998). In these approaches, then, what is relevant is whether the subject considers that the proposition is true in the actual world or not. This contrast captures the intuition that a subject might have contradictory desires but not contradictory beliefs.3 In Giorgi & Pianesi (1997), a different perspective is taken. This proposal adopts Kratzer's (1991) semantics for modality to characterize one of the factors that trigger the subjunctive mood in Romance: these are contexts that require a non-null ordering source. They claim that the possible contexts of evaluation for propositions can be characterized with a scale that ranges from totally realistic to totally non-realistic contexts. The subjunctive/indicative distinction would make it possible to divide this scale into two parts, where the subjunctive would indicate the non-realistic end of such a scale. Crosslinguistic variation would be a result of different possible divisions of that scale. Giorgi and Pianesi claim that their approach is a more fine grained development of the traditional realis/irrealis opposition. Finally, Portner (1997) proposes for Italian, that the indicative mood is used with a modal force of necessity and a prototypically factive modal context, and
3 This intuition, however, does not seem to extend to other predicates that select the indicative in most Romance languages, e.g., promise, suspect, guess, conclude.
230
ELISABETH VILLALTA
that the subjunctive is used whenever the indicative is inappropriate. He, as well, considers his approach as 'basically following the traditional view of indicative as realis mood, indicating truth in the actual world, with the subjunctive mood used when the indicative is inappropriate' (Portner 1999, p.7).4 What the above approaches share, is that they classify contexts as being more or less realistic in some extended sense. Emotive factive predicates (cf. (4)), which select the subjunctive mood in Spanish, are problematic for the realis/irrealis opposition. A sentence containing a verb such as regret presupposes that the subject believes the proposition to be true in the actual world. Moreover, it usually presupposes that the complement is true in the actual world. The approaches above cannot account for the subjunctive mood in this context in a straightforward way. Under Farkas' proposal, we expect these predicates to select the indicative mood since the subject believes the embedded proposition to be true in the actual world. Under Giorgi & Pianesi's approach, we expect the context of evaluation to be realistic, since the embedded proposition is presupposed to be true in the actual world. Both these approaches refer to the emotive character of these predicates to account for their behavior in a language such as Spanish. In particular, Giorgi and Pianesi propose that these predicates introduce an ordering source and thus contribute in turning the context of evaluation into a non-realistic one. Finally, under Portner's proposal, for an emotive factive predicate the modal context of the proposition is expected to be factive. He suggests that these predicates do not select propositions but rather take events as their argument. The indicative would thus not be expected, since the complement would not denote a set of worlds. In what follows, I depart from the view that some feature related to the realis/irrealis notion plays a relevant role. By examining the semantic properties of desire predicates and emotive factive predicates, I will point to a common property that is independent of the realis/irrealis character of the evaluation context of the embedded proposition, namely, they require comparison of alternative propositions. In the following section, my point of departure is then to examine the semantic properties that emotive factive predicates share with desire predicates.
Note that a number of predicates in Italian select the indicative mood even though they do not seem to be prototypically factive, such as suspect, conclude, understand, agree and promise.
COMPARATIVE SEMANTICS FOR SUBJUNCTIVE MOOD
231
2. What do emotive factive predicates share with desire predicates? In this section, I present Heim's (1992) analysis of desire predicates which builds on Stalnaker's (1984) insight that every desire report contains a hidden conditional. I then argue that this conditional semantics should be extended to all predicates that select the subjunctive mood in Spanish.5 We begin with the semantics for propositional attitude predicates adopted since Hintikka (1969). The definition for the verb believe is given in (5). (5)
'a believes φ' is true in w iff: V w' Є Doxa(w): φ is true in w'. (Doxa(w) contains all the worlds that are compatible with what a believes to be true in the world w.)
Following the Hintikka-style analysis, the verb want can be defined as in (6). (6)
'a wants ' is true in w iff: V w'Є Bula(w): φ is true in w'. (Bula(w) contains all the worlds that are compatible with what a wants in w.)
This definition, however, cannot appropriately capture all the characteristics of the verb want. Consider the example in (7), from Heim (1992). (7)
I want to teach Tuesdays and Thursdays next semester.
Heim notes that (7) can be true in a situation in which I would prefer not to teach at all, i.e. in a situation in which in all the worlds that are compatible with what I desire, I do not teach. The definition in (6), however, predicts (7) to be false in this situation. The definition of want thus needs to make explicit that the subject only takes into consideration those worlds that she/he believes are possible. Heim (1992) proposes a new definition for want that captures Stalnaker's insight that every desire report employs a hidden conditional: 'to want something, is to prefer something to certain relevant alternatives, the relevant alternatives being those possibilities that the agent believes will be realizable if he doesn't get what he wants.' (Stalnaker 1984, p.89).
5 Giannakidou (1998) and Portner (1999) already point out that Heim's conditional semantics for these verbs is relevant for mood selection.
ELISABETH VILLALTA
232
Following Lewis (1973) and Stalnaker (1968), a conditional if φ, Ψ is true in a world w iff Ψ is true in all φ-worlds maximally similar to w. Adopting such a conditional semantics, Heim defines the verb want as in (8). (8)
'a wants φ' is true in w iff: For every w' G Doxa(w): Every -world maximally similar to w' is more desirable to a in w than any - -world maximally similar to w'.
Under this definition, for (8) to be true, the following has to hold: each doxastic alternative in which I teach Tuesdays and Thursdays is more desirable than minimally different worlds in which I do not. And each doxastic alternative in which I do not teach Tuesdays & Thursdays is less desirable than minimally different worlds in which I do. In Heim (1992) the formal definition of this verb is stated in a Context Change Semantics. Since, in this paper, I do not use this framework, I present the equivalent definition in a non-dynamic semantics. Heim encodes the relation of comparative similarity among worlds with a family of selection functions: for each world w, there is a selection function Simw from propositions to propositions which maps each p to the set of p-worlds maximally similar to w. (9)
Sim w (p)={w'ep and w' resembles w no less than any other world in p}
Additionally, Heim uses an abbreviation for the ranking of possible worlds in terms of desirability. She introduces
For any w, w', w " Є W, w ' < a , w w " iff w' is more desirable to a For any w Є , p, w"Є q.
W, q W, p
The formal definition of the verb want can now be stated as in (11). (11) [[want]](p)(a)(w) = 1 iff
w' Є Doxa(w): Simw.(p)
COMPARATIVE SEMANTICS FOR SUBJUNCTIVE MOOD
233
In other words, in (11), a wants p is true in w iff for every world w' compatible with a's beliefs, the p-worlds closest to w' are more desirable to a in w than the non-p-worlds closest to w'. Heim notes that, when extending such a definition to other desire predicates, one soon runs into problems. The sentence in (12), from Heim (1992), illustrates the case of the verb wish. (12) John wishes he would teach on Tuesdays. This sentence can only be judged to be true in a situation in which John does not believe that he is teaching on Tuesdays. Since the worlds compatible with his beliefs do not include worlds in which he teaches on Tuesdays, the definition in (11) cannot be adopted. Heim suggests that, in this case, the proposition is evaluated with respect to the revised set of worlds compatible with his beliefs. We add to the set of his doxastic alternatives all the (maximally similar) worlds in which John teaches on Tuesdays. The meaning of (12) can then be paraphrased as 'John thinks that if he taught on Tuesdays he would be in a more desirable world then he is in (because he is not teaching on Tuesdays).' Heim's definition of the verb wish is given in (13). (13) [[wish]](p)(a)(w)=l iff V w' e revised Doxa(w):Simw,(pK!W Simw,(-p) Heim (1992) furthermore extends her proposal to factive desire predicates such as be glad, illustrated by the following example. (14) John is glad that he will teach on Tuesdays. The sentence in (14) has the presupposition that John believes in the truth of the complement. Here as well, since John's doxastic alternatives only include worlds in which he teaches on Tuesdays, the definition of the predicate be glad has to make reference to a revised set of his doxastic alternatives. The meaning of (14) can then be paraphrased as 'John thinks that (because he will teach on Tuesdays) he is in a more desirable world than he would be if he would not teach on Tuesdays. ' The definition of be glad is given in (15). (15) [[be glad]](p)(a)(w)=l iff V w' G revisedDoxa(w):Simw,(pK5W Simw,(-p)
234
ELISABETH VILLALTA
To summarize, desire predicates such as wish, want and be glad share their core semantics, with the only difference that wish and be glad make reference to the revision of the set of worlds compatible with the subjects' beliefs. After this brief presentation of Heim's semantics for desire predicates, we now return to the class of predicates that select subjunctive mood in Spanish. The desire predicates just discussed all select the subjunctive mood in Spanish. Crucially, whether counterfactual worlds are evaluated or not does not play a role for the selection of subjunctive mood. The conditional semantics adopted so far can be straightforwardly extended to all other predicate classes that select the subjunctive mood, such as directives, predicates of doubt, modals and emotive factives. I assume that the comparison relation (or scale) is contributed by the lexical meaning of each predicate, since not all predicates require a relation of desirability, as for example be surprised, doubt and other predicates. To conclude, I present a first attempt for a generalization in (16). (16) Preliminary generalization for subjunctive mood in Spanish: An embedded proposition p requires the subjunctive mood, if p is compared to non-p on a scale introduced by the matrix predicate. 3. Subjunctive Mood as Comparative Mood In this section, I argue that the proposed semantics has to be revised and that it crucially involves comparison of the embedded proposition p with contextual alternatives. After presenting evidence from contexts that contain more than two alternatives, I develop a new analysis for these predicates. I then restate the generalization for the predicates selecting the subjunctive mood in Spanish. 3.1 Comparing more than two alternatives We now examine a context that provides more than the two alternatives p and - p . It illustrates that the definition of want adopted so far really only characterizes a special case. Consider the scenario in (17). (17) Sofía has promised to bring a desert to the picnic. Victoria knows that there are three possibilities. She could prepare a chocolate cake, even though Victoria considers that completely unlikely because it represents far too much work. She might bring an apple pie, which Victoria thinks is very likely since she can just buy it at the bakery nearby. Or she might bring ice-cream, which seems most likely to
COMPARATIVE SEMANTICS FOR SUBJUNCTIVE MOOD
235
Victoria, since she usually has some in her freezer. Victoria prefers the chocolate cake over the apple pie and the apple pie over the ice-cream. In this situation, the sentence in (18) is intuitively judged to be false. (18) Victoria wants Sofía to bring the apple pie. However, if we apply Heim's definition of the verb want, the sentence in (18) comes out as true in this scenario: the worlds in which Sofía brings the apple pie are more desirable to Victoria than all the minimally different worlds in which she does not. This is true, since this set does not contain worlds in which she brings chocolate cake (these are not minimally different, because Victoria considers them completely unlikely). The worlds in which she does not bring apple pie are less desirable to Victoria than minimally different worlds in which she does. These contain worlds in which she brings ice cream. Hence, a conditional semantics does not make the correct prediction for this scenario. I conclude that the semantics of these predicates involves comparison of p with its contextual alternatives rather than with - p . In the following section, I propose to revise the semantics accordingly. 3.2. Comparative predicates Following the argument presented in the previous section, I propose to revise the definition of the verb want as in (19). Its lexical entry is given in (20). (19) 'a wants φ' is true in w iff ψ that is a contextual alternative to φ: φ
q e g(C): p
where
236
ELISABETH VILLALTA
In (20), the verb want carries an index that stands for a variable anaphoric to a contextually determined set of propositions (it is an index of type « s , t > , t » ) . This variable receives its content from the variable assignment g. I may mention here, that this kind of variable has also been used before for similar purposes, such as for the domain of quantification of only in Rooth (1992) and the resource domain of adverbs of quantification in von Fintel (1994). A sentence a wants p is then true in w iff for all propositions q in a contextually determined set C, p is more desirable to a in w then q. The semantics proposed for want can be extended to all predicates that select subjunctive mood in Spanish. The following generalization for subjunctive mood in Spanish can now be presented. (21) New Generalization for Subjunctive Mood in Spanish: An embedded proposition p requires the subjunctive mood, if p is compared to its contextually available alternatives on a scale introduced by the matrix predicate. Note that this characterization of the licencing context for subjunctive propositions is closely related to recent proposals for polarity items. In Israel (1996), polarity items are scalar operators that require a scalar model (structured set of propositions) in the context. In Krifka (1995), negative polarity items introduce ordered alternatives into the semantics. As is well known, polarity items and subjunctive clauses have a strikingly similar distribution (cf. Nathan and Epro 1984, Giannakidou 1998), and a parallel semantics would of course be a welcome result. We are now in the position to divide predicates into two semantic classes. The predicates that select the indicative mood in Spanish fall under the class of acceptance predicates (cf. Stalnaker(1984), 'to accept a proposition is to treat it as a true proposition in one way or another - to ignore for the moment at least, the possibility that it is false', p.79). The Hintikka-style definition of believe requires that the embedded proposition be added to the derived context set (the worlds compatible with the subjects beliefs) and that all other alternatives be eliminated from this derived context. I propose that this characterizes all predicates that select the indicative mood in Spanish. The second predicate class, predicates that select the subjunctive mood, requires the subject to compare the embedded proposition p to its alternatives on a scale. I will call this class of predicates comparative predicates.
COMPARATIVE SEMANTICS FOR SUBJUNCTIVE MOOD
237
Bolinger (1968), Terrell & Hooper (1974) and others have proposed a divide that intuitively seems similar. They claim that propositional attitude predicates can be categorized into assertives and non-assertives. They point to the fact that the predicates requiring the indicative mood in Spanish correspond to the class of assertive predicates. Their definition of assertive predicates however differs from the definition of acceptance predicates. They claim that assertive predicates require the speaker to assert the embedded clause, while non-assertive predicates require the speaker to presuppose the embedded clause or neither assert nor presuppose it. This characterization thus differs from the one presented here. We still need to show that characterizing the class of predicates selecting the subjunctive mood with an alternative semantics correctly excludes all predicates that select the indicative mood. For most predicates of this class, such as declarative predicates, predicates of certainty, commissives and fiction verbs, it seems obvious that such an alternative semantics is not appropriate. For example, to say thatp cannot mean to say p more than its alternatives, simply because what is said cannot be graded. However, nothing seems to prevent us from attributing an alternative semantics to a verb such as believe, even though it selects the indicative mood. To believe that p could mean to consider p more believable/likely than its alternatives. Interestingly, the verb believe displays crosslinguistic variation for its mood selection properties. In Italian, for example, believe selects the subjunctive mood contrary to Spanish. I propose then that believe can in principle receive slightly different interpretations and that in a language in which believe selects the subjunctive mood, it does not have the same meaning as in a language in which it selects the indicative mood. Crosslinguistic variation would then indicate that the predicates have slightly different meanings in the languages where mood selection differs. We expect little crosslinguistic variation with predicates such as say, know, understand, conclude, where it is clear that comparison of alternatives cannot be part of their meaning. An interesting fact is that predicates for which comparison of alternatives is most salient, such as prefer and wish, select the subjunctive mood in most languages. 4. Empirical eviden ce for the proposal In this section, I provide some empirical evidence that these two classes of predicates form natural classes and that a characterization in terms of
238
ELISABETH VILLALTA
comparison of alternatives of the predicates selecting subjunctive mood is justified. 4.1 Degree Modification Predicates that select the subjunctive mood allow more readily for degree modification than predicates that select the indicative mood. (22) illustrates the indicative mood, while (23) illustrates the subjunctive mood. Degree modification with 'mucho' ( 'a lot') is only possible in (23). (22) *Violeta concluye /cree /sabe mucho que no es posible. Violeta concludes/believes/knows a-lot that not is possible. "Violeta very much concludes/believes/knows that it is not possible." (23) Violeta quiere/espera/duda mucho que sea posible. Violeta wants/hopes/doubts a lot that is (Subj.) possible. "Violeta very much wants/hopes/regrets that it will be possible." The fact that the predicates selecting subjunctive mood more readily allow for degree modification provides linguistic evidence for the claim that these predicates compare alternatives on a scale. 4.2 Correct Attitudes Stalnaker (1984) proposes that a prepositional attitude concept is an acceptance concept if the attitude is said to be correct whenever the proposition is true. This criterion allows us to contrast the two classes of predicates under discussion. (24), with predicates that select the indicative mood, is possible, while (25), with predicates that select the subjunctive mood is not.6 (24) What she said/affirmed/was certain about/believed/ realized/ understood/dreamed turned out to be correct. (25) *What she desired/advised/regretted turned out to be correct. A Hintikka-style semantics for predicates selecting the indicative mood thus makes the correct predictions.
Predicates of doubt and probability, however, do not pattern with the other predicates that select the subjunctive. I leave a more detailed analysis of these predicates for future research.
COMPARATIVE SEMANTICS FOR SUBJUNCTIVE MOOD
239
4.3 Preposed complements Bolinger (1968), Terrell & Hooper (1974), and others point out that the class of predicates which permit their complements to be preposed in English {assertive predicates) correspond to the class of predicates that selects indicative mood in Spanish. In particular, Bolinger (1968) provides extensive evidence for this claim. The predicate in (26) selects the indicative in Spanish and allows preposing of the complement, while the predicate in (27) selects the subjunctive mood in Spanish and disallows preposing. (26) I believe they're ready. (27) I'm sorry you lost it.
They're ready, I believe. *You lost it, I'm sorry.
To conclude, I have provided empirical evidence that indicates that the two classes of predicates form indeed natural classes. 5. Subjunctive mood in other contexts We now turn to a few other contexts in which the subjunctive mood appears in Spanish to see whether the approach adopted so far can be extended to these. We mention here the cases where subjunctive mood appears in the antecedent of a conditional, in questions, and is triggered by negation under certain circumstances. A number of studies agree that in these cases there is an implicature concerning the speaker (cf. Rivero 1971, Bell 1980, Borrego et al 1985, Sastre 1997 and others). The subjunctive would trigger the implicature that the speaker considers non-φ more likely than φ. For example, in (28) the speaker is more inclined to believe that Ana will not come. If the indicative were used instead, this implicature would not arise. (28) Si Ana viniera a. regar las plantas, no se morirían. If Ana comes (Subj.) to water the plants, not SE die (Cond.). "If Ana would come to water the plants, they wouldn 't die. " Similarly, in (29), an example from Sastre (1997), the speaker considers more likely that she did not say yes. (29) Crees que haya dicho que sí? Think that has (Subj.) said that yes? "Do you think that she said yes?"
240
ELISABETH VILLALTA
Finally, in (30), an example from Rivero (1971), the speaker is more inclined to believe that the Belgian did not win the race. (30) Los corredores no creen que el belga ganara la carrera. The runners not believe that the belgian won (Subj) the race 'The runners don't believe that the Belgian won the race." As before, it is possible to show that the generalization 'the speaker considers less likely than non-φ' is in fact a special case of 'the speaker considers φ less likely than other contextually available alternatives'. Similarly to what I presented before in section 3, this could be done with examples that involve more than two alternatives. For reasons of space, I will not illustrate it here. To conclude, the intuition that speakers have for sentences (28)-(30) suggests that these contexts share with the embedded contexts discussed before that contextually available alternatives are being compared. The details of an analysis of these cases remain to be made precise. 6.
Conclusion I have proposed that, in Spanish, subjunctive mood is selected by predicates that express comparison of alternatives. Whether these alternatives are real or counterfactual is irrelevant for this classification. I then depart from the idea that the extent to which the context corresponds to reality plays a significant role for the distribution of subjunctive mood. At present, I have provided an analysis of subjunctive mood embedded under propositional attitudes. Ultimately, my goal is to determine a unified account for all other contexts in which subjunctive mood appears.
REFERENCES Bell, Anthony 1980. "Mood in Spanish: A discussion of some recent proposals". Híspanla 63.377-390. Bergen, John 1978. "One rule for the Spanish Subjunctive". Híspanla 61.218234. Bolinger, Dwight 1968. "Postposed Main Phrases: An English rule for the Romance subjunctive". Canadian Journal of Linguistics 10.125-197.
COMPARATIVE SEMANTICS FOR SUBJUNCTIVE MOOD
241
Borrego, J., J.G. Asencio & E. Prieto. 1985. El subjuntivo. Madrid: Sociedad General Española de Librería. Farkas, Donka 1985. Intensional Descriptions and the Romance Subjunctive Mood. New York: Garland. .1992. "On the Semantics of Subjunctive Complements". Romance Languages and Modern Linguistic Theory ed. by Hirschbuehler & Koerner.69103. Amsterdam & Philadelphia: John Benjamins. Giannakidou, Anastasia. 1998. Polarity Sensitivity as (Non) Veridical Dependency. Amsterdam & Philadelphia: John Benjamins. Giorgi, Andrea & Fabio Pianesi 1997.Tense and Aspect, From Semantics to Morphosyntax. Oxford: Oxford University Press. Heim, Irene 1992. "Presupposition Projection and the Semantics of Attitude Verbs". Journal of Semantics 9.183-221. Hintikka, Jaakko 1969. Models for Modalities. Dordrecht: Reidel. Israel, Michael 1996. "Polarity Sensitivity as Lexical Semantics". Linguistics and Philosophy 19.619-666. Klein, Philip Walter 1974. Observations on the Semantics of Mood in Spanish. Ph.D. dissertation. University of Washington. Kratzer, Angelika 1991. "The Notional Category of Modality". Words, Worlds, and Contexts: New Approaches in Word Semantics ed. by Eikmeyer, H. J. & Rieser H. 38-74. Amsterdam: de Gruyter. Krifka, Manfred 1995. "The Semantics and Pragmatics of Polarity Items". Linguistic Analysis 25.209-257. Lewis, David 1973. Counterfactuals. Cambridge: Harvard University Press. Nathan Geoffrey S. and Epro Margaret W. 1984. "Negative Polarity and the Romance Subjunctive". Papers from the Xllth Linguistic Symposium on Romance Languages ed. by Baldi P. 517-529. Amsterdam & Philadelphia: John Benjamins. Palmer, Frank R. 1986. Mood and Modality. Cambridge: Cambridge University Press. Portner, Paul 1997. "The Semantics of Mood, Complementation, and Conversational Force". Natural Language Semantics 5.167-212. . 1999. "The Semantics of Mood". Glot International Vol 4.3-9. Quer, Josep 1998. Mood at the Interface. Ph.D. dissertation, University of Utrecht. Rivero, María Luisa 1971. "Mood and Presupposition in Spanish". Foundations of Language 7.305-336.
242
ELISABETH VILLALTA
Rooth, Mats 1992. "A Theory of Focus Interpretation". Natural Language Semantics 1.75-116. Sastre, María A. 1997. El subjuntivo en español. Salamanca: Ediciones Colegio de España. Stalnaker, Robert 1968. "A Theory of Conditionals", Studies in Logical Theory, American Philosophical Quarterly, 2.98-112. -. 1984. Inquiry. Cambridge, Mass.: MIT Press. Terrell Tracy & Hooper, Joan 1974. "A semantically based analysis of Mood in Spanish". Hispania 57.484-494. von Fintel, Kai 1994. Restrictions on Quantifier Domains. Ph.D. dissertation, University of Massachusetts, Amherst, Mass.
INDEX OF TERMS & CONCEPTS A Adjectives, 3, 11, 13-14, 16, 1819,27-40, 106, 108, 111-116, 118, 120, 179-180, 182,212, 219,221 placement, 38, 106, 108, 112, 120 Adverbs, 5, 7, 9-16, 18-19, 21-23, 99, 139, 173, 186-187, 190, 192-193,236 higher, 5, 11-12, 15 lite, 5-6, 9-10, 16-23 lower, 15-16 scalar, 10, 12, 15-16, 18-20 Agree, 49, 51-52 Agreement, 3, 50-52, 54, 89, 99, 106, 108, 110-118, 153, 158159, 163-164, 173-175, 179180,182 A-movement, 153 Anaphor, 7, 161-163, 172-173, 175-177, 180-181, 184-186, 193 S(implex) E(xpression) anaphor, 171,182 Anti-agreement, 50 Assertive predicates, 237, 239 Assertives, 237 non-assertives, 237
Binding, 2, 153, 161-163, 177, 180, 182, 184-186,223
Case, 2, 4, 8,43-44, 51,53,56, 84-86,91-92,94, 102-103, 118119, 157-161, 167, 172-175, 177, 180, 182-186, 192-193 c-commanding antecedent, 154155 Cleft, 62-64 Clitic, 2, 7, 9, 12,45,47-48,5054,62,86,92,94, 120, 171-173, 175-176, 181-187, 191-193,210 reflexive, 7, 85, 169-174, 176, 180,182, 185, 190-191, 194 Co-composition, 215-216, 221222, 224 Conditional, 39, 187, 227, 231232,234-235,239 Conditional semantics, 227, 231232, 234-235 Conjunction introduction, 29-30, 34 Constraint Demotion Algorithm, 2-3, 122-123, 129-130, 132 Constraint interaction, 122-123 Constructionism, 111 Control, 2, 153-154, 157-159, 161, 164-166, 177, 179-180, 182 adjunct, 2, 153, 159 Arbitrary control PRO, 153 non-obligatory control, 2, 153, 165 object, 158 obligatory control, 2, 153-154, 164-165 PRO, 154, 161-162, 164
244
INDEX OF TERMS & CONCEPTS
Control by the matrix subject, 2, 153 Control by the object, 2, 153 Co-ordination, 142, 148 Co-referential, 156, 162 D Definite NPs, 61 Desire predicates, 6, 228, 230-231, 233-234 Determiner Phrase, 1-3, 27, 106 Developmental stages, 3, 121-122 Discloser, 7, 173, 190, 192-194 Disclosure of the impersonal, 178, 191 Discourse, 2, 30, 35, 61, 66-67, 95, 102, 122-124, 126, 128, 132, 134, 139, 151, 163, 197-199, 202,204,208,219,221-222 Discourse topic, 124 Dominance, 132 do-support, 54 Doubling, 2, 4, 45, 48, 50-54, 8286, 120, 191 DP-movement, 2, 157 E Echo-questions, 61 ECP, 2, 45, 54 Ellipsis, 56, 155 Embedded clauses, 59, 227 Emotive factive predicates, 230 Entailment, 29, 144, 146 Exclusion operators, 142, 147-148 F Feature checking, 157, 183 Feature-movement, 58
Features Case, 158-161, 183 EPP, 157-158, 160 morphological, 109 strong, 58, 108-109 Focus, 50-51,55-56, 63, 137, 147148, 151,208,222, 242 Free linking, 89, 97 Full Access, 8, 32,41, 111, 120, 123, 135 Full Functional Hypothesis, 3, 106, 109-110 Full Transfer, 8,32-33,41, 111, 120, 123, 135 Functional application, 29-30 Functional category, 3, 27, 106, 109, 116, 173, 181, 183, 190, 193-194 G Generative lexicon, 212 H Head-final, 4, 73-74, 76-77, 79-83, 85-86 Head-initial, 4, 73-74, 76-78, 8083,85-86 Head-initial & head-final, 4, 74, 82-83, 85-86 Head-movement, 2, 53-54, 58 Heavy-APs, 33
I Impersonal construction, 183, 194 Impersonal se/si/się, 169, 173, 178-179, 185, 193 Indefinite pronoun, 7, 169, 172173, 180, 182, 185-186, 190, 193
INDEX OF TERMS & CONCEPTS
Indicative, 38, 227-230, 236-239 Infinitives Inflected infinitives, 2, 153, 154-155, 158, 160-161, 163165 non-inflected infinitives, 2, 153157, 161, 164-165 Inflectional morphology, 2, 153154, 163-165 Information structure, 126, 134, 214 Input, 2, 5, 27-28, 32-34, 40, 76, 81, 110, 122-125, 129, 131, 184-185, 197-199,201,203204,207-210 Input modifications, 198 Insertion covert, 58-59 Intensional, 59, 65-69, 229 Interpretation, 1, 3, 7, 27, 29-31, 34-39,41,47-49,66-67,71,89, 97-99, 101-102, 107, 134, 141, 144-145, 155-156, 163-165, 184-185, 189,215 Exhaustive, 67 Interrogative inflection, 50 Intrinsic existential force, 172, 186-187, 192 Inversion, 2, 43-44, 47-50, 53, 5556 complex, 2, 47-49, 51-55 Irrealis, 6, 50, 227-230 L Learnability, 34, 53 Learning, 3, 84-85, 110, 112, 121124, 128-129, 131-132, 134, 208 Lexical representations, 214-215 Logical types, 29-30
245
M Merge, 59, 159-160 Minimal Link Condition, 159-160 Modality, 228-229 Move, 49, 54, 65-67, 157, 159-161 N Negative polarity items, 236 Non-intensional, 59, 65, 67, 69 Noun-movement, 3-4, 29, 31-33, 35,38-39
Optimality Theory, 2, 8, 122, 135 Optionality, 1,57,70, 133-134 Ordinals, 148, 150 Overt pronoun, 127, 162 P Parameter, 4, 27, 41,49, 51-52, 74,82-83,86, 106, 108, 111, 117, 129 Parametric change, 32 Parametric model, 82 Parametric variation, 1, 3, 7, 106, 108, 119, 169, 173, 188, 190, 194 Phase, 44, 194 Postsyntactic composition, 7, 89, 97 Presuppositionality, 61, 64, 68-70 pro, 6, 52, 155-156, 161-162, 164165, 183, 193,227,229-230, 232,236 Procrastinate, 158-159 Pronoun binding, 162 Propositional attitude predicates, 6,227,229,231,237 Punctual transition, 217, 219
246
INDEX OF TERMS & CONCEPTS
Q Quantification Cardinal quantifiers, 142, 144, 147 Quantification over events, 5, 137, 143 Quantificational variability, 71, 186-187, 193 R Redundancy, 198 Reflexivity theory, 177, 181 Robust Interpretive Parsing, 123 S Second language acquisition, 1-2, 4,8,40, 121-122, 197,209 Semantic variation, 169, 188 Semantics possible world semantics, 229 Sloppy reading, 155 Specificity constraint, 66 Split antecedents, 156-157 Strong forms, 9 Structure Building Hypothesis, 3, 106, 109 Subject Initiation Principle, 7, 98 Subjects arbitrary plural subjects, 165 non-overt, 153 null-subjects, 2, 121-122, 125, 128 Subjunctive, 1, 6, 227-231, 234, 236-240 Summation, 137-140 Superlatives, 148, 150 Syntactic modifications, 199
T Temporal adverbials, 5, 137, 139, 141, 151 Temporal frame, 138-141, 146, 150 Temporal location, 5, 139, 141, 150-151 Temporal locator, 140 Temporal measure adverbials, 142, 146 Tense, 2,43-44, 51,53, 100-101, 241 Typological shift, 73-74, 85 U Unaccusative mismatches, 7, 8990,95-96, 102 Unaccusativity, 89-90, 102 Unselected Datives, 172, 186, 188, 190 W Weak forms, 9, 13,63 Weak/strong distinction, 5, 12 Weight, 16,23-24 which-phrases, 69 wh-in situ, 1, 58-59, 63-64, 68-69 Word order, 1,3-5,9, 17,23,27, 32-33, 35, 40, 73-74, 76, 79-80, 82-83,85,87,91, 109, 111, 115, 198-199,209-210 major, 82, 134, 142, 172, 199 minor, 82 Word order shift, 4, 73